Changelog

What has changed

Every change to Time Trakkr, newest first: new features, improvements and fixes. Each note says what changed, and where it matters, what it will not do.

November 2026

New

Map several repositories to a project at once

Under Settings → Agent time → Repositories, each repository in the GitHub list now has a box to tick, and Select all on this page ticks a whole page. Pick the project and task once in the bar that appears, and every ticked repository is mapped to it. Ticks are kept while you search and page, so one client's repositories can be gathered from several searches.

Up to 100 are mapped in one step, and it is all or nothing. A repository that is already mapped can be ticked too: the bar says so first, and its rule for every branch moves to the new project and task. Branch rules are not changed, and agent time already recorded stays where it is.

The assistant can map several repositories at once as well.

Improved

Agent time is free on every plan

Agent time has never been a paid feature, but several pages called it an "add-on", which read as something you pay for. They now say what is true: it is free on every plan, and an admin switches it on under Settings → Add-ons. Nothing about how it works or what it costs has changed, and the settings page keeps its name.

The home page has also been rewritten to say in one paragraph what Time Trakkr is, who it is for and what it costs, with answers to whether it is free, how it compares to Harvest, whether it monitors screens (it does not) and whether it can bill AI agent time.

New

Choose the task for an agent session from inside it

A session's project and task used to come only from the workspace's repository rules or a .timetrakkr.json file, and the only way to put it somewhere else was to change the task on the pending row. Now you can choose it while you work: /timetrakkr:task acme bug in Claude Code, /timetrakkr-task in Cursor, the timetrakkr-task skill in Codex, or timetrakkr task acme bug in a terminal. It searches the projects and tasks you can log time to.

Once a session has recorded time, you choose what moves: the whole session, or only the time from now on. A day already accepted stays where it is. Time from now on goes on its own row in your pending list, as a new part of the same session. The choice lasts for that session only.

A chosen task also tracks a session that no rule covers, including one in a folder that is not a git repository, without sending the repository's address or branch. It will not track a branch whose rule is for a project you cannot log time to, and time on a chosen task is never accepted automatically.

The agent never chooses a task for you. The command is for you to run, and the agent is told to ask you, not to pick, whenever there is a choice to make.

This needs version 1.3.0 of the plugins, or of @timetrakkr/cli.

Fix

The API answers a malformed id as "not found" instead of a server error

Asking the API for something by an id that is not a valid UUID, such as GET /api/v1/projects/fda7bc05-0000/user_assignments, used to fail with a server error. It now gets exactly the answer a record that does not exist gets on that endpoint: the same status, usually 404, and the same message.

The same goes for id filters on lists. GET /api/v1/time_entries?project_id=…, /api/v1/invoices?client_id=… and /api/v1/expenses?user_id=… with an id that is not a UUID now return an empty list, as /api/v1/projects?client_id=… already did.

Nothing that worked before changes. A request with a valid id gets the same answer it always did.

Fix

API payment deletes check the invoice, and list filters refuse unknown values

DELETE /api/v1/invoices/{id}/payments/{paymentId} read only the payment id: it would delete a payment that belonged to a different invoice from the one in the path, and answered 200 for a payment that did not exist. It now answers 404, and deletes nothing, when the invoice does not exist, the payment does not exist, or the payment belongs to another invoice.

Recording or removing a payment through the API now updates the invoice's state the way the app does, so an invoice paid in full reads paid, and one whose payment is removed goes back to open or partially_paid. Recording the payment that settles an invoice now sends the invoice.paid webhook, as it does from the app.

GET /api/v1/invoices?state=, GET /api/v1/users?role= and GET /api/v1/time_entries?approval_status= now answer an unknown value with a 400 that names the parameter and lists the values it takes. Before, the first two failed with a server error and the third ignored the filter and returned every entry.

Fix

API time reports filter by client, project, engagement and person

GET /api/v1/reports/time/… ignored client_id and engagement_id and returned the whole workspace's report as if it were filtered. Time reports now take client_id, project_id, engagement_id, user_id and billable, and every API report answers a parameter it does not take with an error that names it, instead of quietly leaving it out. Rows of GET /api/v1/reports/time/engagements now carry engagement_id beside engagement_name (null for time on projects in no engagement), so two engagements with the same name can be told apart.

GET /api/v1/projects/{id}, and the response to creating or updating a project, now include the client's name. Creating a time entry through the API now answers with the person's, project's and task's names, the same as reading the entry back.

In the assistant and over MCP, the budget answers now include each project's id and its client's id, so a follow-up question about one project does not have to look it up again by name.

Fix

Tasks that need a Jira issue hold for copied rows and agent time

A task that needs a Jira issue on every new entry now holds in three places it did not.

Copying rows. "Copy rows from last week", and copying a day's rows at zero hours, used to add an entry with no issue on such a task. Those rows are now left out, and the copy names the task so you can add that entry with its issue. Copying a day with time and notes was already checked.

Agent time. Agent time never carries a Jira issue, so it is no longer accepted onto a task that needs one, whether automatically or by you. It waits in your pending list with the reason: accept it onto another task, or discard it. An accepted agent entry cannot be moved onto such a task either.

The switch. Needs Jira on a project's tasks can be ticked only once someone in the workspace has connected Jira. Without a connection every entry on the task would be refused. A task already switched on keeps the rule, and the page warns you while Jira is disconnected.

New

A project's start date and planned end date appear on its page

A project's dates were saved under Edit project and then shown nowhere. They now appear on the project page, under its name, in your workspace's date format. Once an active project is past its planned end, the page says so.

The fields are now called Start date and Planned end date. Both stay optional, and neither stops anyone tracking time before or after them.

The assistant can read both when it looks up a project, and GET /api/v1/projects returns them as starts_on and ends_on, the same names POST and PATCH already took.

New

The changelog is now public

The release notes you see under Settings → What's new are now also on the public website, at timetrakkr.com/changelog. Anyone can read them without signing in, which makes it easier to point a colleague or a prospective customer at a change.

Each note has its own page, and a link straight to its place in the full list, so you can share exactly the change you mean.

What it will not do: it does not show notes for add-ons that have not been released yet, or notes dated ahead of today. Notes about the website itself, such as new blog articles, stay out of it, as they do in the app.

Fix

A manager's API token no longer reaches invoices and estimates, and organization keys are admin-only

In the app, invoices and estimates are for owners and admins. The REST API was looser: a manager's personal access token could list and read every invoice and estimate, and download an invoice as a UBL e-invoice. It now matches the app.

This changes what existing integrations get. If an integration uses a manager's token to read invoices or estimates, those requests now answer 403, with the usual {"message": "…"} body, instead of the data. Every /api/v1/invoices and /api/v1/estimates endpoint behaves this way, reading and writing alike. A manager's token was already refused when it tried to create, change, send or delete an invoice or estimate; reading is what is new.

What does not change:

  • An admin's or owner's token works as before.
  • An organization key works as before: with read access it reads invoices and estimates, and with write access it can also create and change them.
  • A member's token was already refused, and still is.
  • The uninvoiced report (/api/v1/reports/uninvoiced) is about time, not invoices, and still answers a manager's token.

To keep such an integration working, have an admin or owner create its token.

Organization keys are now managed by owners and admins only. An organization key acts for the whole workspace, invoices included, so a manager who could create one could get around the rule above. On the Developer page a manager now sees a note in place of the keys table, and creating, listing or revoking an organization key is refused for anyone below admin. A manager's own personal access tokens are untouched.

Keys a manager created before this change keep working: nothing was revoked. An owner or admin can see them on the Developer page and revoke any that should not exist.

The API reference and the OpenAPI document now also list these refusals, and the answers a member's token has received since October: 404 for a project the member is not assigned to or for another person, and 422 for a time entry whose project, task or person is not in the workspace.

Improved

Members see only the tasks and clients of their own projects

In a workspace that restricts members to their assigned projects, a member's task lists now follow the same rule as their project picker: the tasks of a project they are not on are no longer listed anywhere, including the assistant's search. The same goes for every way the assistant can be asked about a project: its budget settings, its roles, its plan against actual hours, budget and engagement reports, logging time or starting a timer. A project the member is not on is answered as if it did not exist, and the budget and engagement reports in the app leave it out too. The list of reports published to clients is now for owners, admins and managers in the assistant, as it already was in the app.

Clients follow suit for every member. A member sees only the clients of the projects they can see, and only each client's name and currency. The address, notes, invoice terms and contact people are kept for owners, admins and managers, in the app, in the assistant, through Claude and other MCP connectors, and over a member's API token. A client with no projects yet is not listed for members.

Owners, admins and managers see exactly what they saw before, and so do organization API keys. Nothing about who can create or edit clients has changed.

Improved

A public status page

status.timetrakkr.com checks Time Trakkr from outside every few minutes: the app itself, and the MCP server that Claude, ChatGPT, Perplexity and other AI clients connect to. When something is wrong, it says so there, along with what we are doing about it, and planned maintenance is posted in advance.

It is linked from the footer of the website, the contact page and the MCP help article. It reports whether the service is reachable; it does not show anything about your own workspace.

Improved

Connect ChatGPT, Gemini CLI and Perplexity to your workspace

Time Trakkr's MCP server already worked with any client that speaks MCP and signs in with OAuth. ChatGPT, Gemini CLI and Perplexity have now each been tested end to end: reading your tracked hours, starting and stopping a timer, and drafting an invoice, which stays a draft and is never sent. The help article on MCP and AI clients has the steps for each one.

ChatGPT and Perplexity need a paid plan to add a custom connector. For Gemini, it is the Gemini CLI that connects; the Gemini web app does not take custom MCP servers.

Fix

Clearer setup steps in the agent plugins and command-line tool

Version 1.2.1 of the plugins for Claude Code, Codex and Cursor, and of @timetrakkr/cli, changes wording only. Nothing about how time is measured or recorded is different.

  • timetrakkr init --cloud now lists every step: create a personal access token with Agent time only and For a cloud environment ticked, add it as the environment's network secret, then start a new session on that environment.
  • When signing out cannot be confirmed on the server, the message now sends you to Integrations → Connected apps, where the sign-in is listed.
  • The plugins' instructions describe cloud agents, Cursor and auto-accept as they work today, and name the real public repository.

Plugins update the usual way for each agent. A cloud setup that pins a version keeps running the one it names until you change --pin.

Fix

Restarting a timer over the API respects a submitted week

PATCH /api/v1/time_entries/{id}/restart could start a timer on an entry in a week that had already been submitted for approval, adding time to a total somebody had signed off. The timesheet already refused this.

The endpoint now refuses it too, with a 409 and the same message. Ask the approver to reopen the week first.

New

The assistant can turn an estimate into a draft invoice

Ask "turn estimate 42 into an invoice" and the assistant does what Accept → invoice and Convert to invoice do on the Estimates page: a sent estimate is accepted and converted, an accepted one is converted, and you get a link to the new draft.

It will not convert a draft or declined estimate, and asking twice never makes a second invoice. It only ever creates a draft: nothing is sent or emailed to your client until you send it yourself from the app.

New

Agents can check in work that has no repository

Until now a check-in needed a git repository that a rule mapped to a project, so agents doing other work, such as client briefings or email, recorded nothing. An agent connected through the Time Trakkr connector can now look the project up and check in against it.

Time Trakkr checks that you can log time to that project and never guesses. A project you cannot log to is refused, and the agent is told which ones you can. The time reads No repository: project chosen by the agent, is never accepted automatically, and is left out when you accept the whole day, so you confirm the agent's choice first.

A token with only Agent time ticked cannot name a project. See agent time for agents in the cloud for the Claude Duet setup.

New

Claude Code cloud sessions can measure their time

A Claude Code cloud session does not load the plugin, so until now it could only check in. Run npx @timetrakkr/cli init --cloud in the repository and commit the file it writes, then give the cloud environment a network secret for your workspace's address holding an Agent time token. Then start a new session, on the environment that has the network secret. The hooks measure each turn and record it as Measured and Cloud.

The token never enters the session: Claude Code's proxy adds it to requests for your workspace on the way out. The hooks do nothing on anyone's own computer, so nobody with the plugin is counted twice.

It does not work in a cloud session with more than one repository, and it is for Claude Code only: Cursor and Codex cloud agents still check in. If the repository also tells the agent to check in, remove that instruction, or the same work is recorded twice. Accepting a whole day leaves any session flagged as a possible duplicate for you to check and accept on its own. See agent time for agents in the cloud.

Fix

Estimates your client accepted can be turned into invoices

When a client accepted an estimate from its link, the estimate was marked accepted but no invoice was made, and the Estimates table offered no way to make one. Only the API could convert it.

Accepted estimates that have not been converted now show Convert to invoice. It makes a draft invoice carrying the estimate's client, subject and line items, and takes you to it. Accepting from the link still does not create the invoice by itself, so nothing is invoiced until you choose to.

Fix

Copied rows need a note before they take hours

Copy rows (no time or notes) makes the same projects and tasks as your most recent day at zero hours, with no notes. In a workspace that requires notes, those rows were a way around the rule: the hours could be filled in later and the entry saved without a note.

Copying the rows still works, and an empty row still needs no note. The note is asked for when a copied row first gets time: entering hours on it, or starting its timer, is refused with the usual "notes are required" message until it has one. That holds on the timesheet, for the assistant and over the API.

Entries that already had hours before notes were made required are not affected. Changing their hours is still allowed.

Fix

Settings and the assistant describe a few things correctly

  • The timer watchdog's second threshold was labeled Tell admins after. It has always told the person's manager, and nobody when they have no manager, so it now reads Tell their manager after.
  • The invoice Configure page said the company name is changed under Settings → General. That page is now called Settings → Preferences.
  • The assistant was told that a review could only be sent from the app, and that the paid plan covered four features when it covers six. It now knows it can send a review, after showing you a preview, and names every paid feature.
  • The Manager role's description on a person's page said managers manage clients and tasks. They manage projects and the tasks on them; clients and the shared task library are for administrators. Nothing about what a manager can do has changed, only the description.
  • The assistant was told it could not approve or return a submitted week. It can, in a session that may make changes, under the same rules as the app.
Fix

Only sent or accepted estimates convert to invoices, everywhere

The Estimates page only ever offered to convert a sent or accepted estimate, but the API would convert any estimate, including a draft nobody had sent or one the client declined. It now refuses those with a clear error, the same rule as the page. A sent estimate is still accepted as it converts, and converting an estimate a second time still returns the invoice it already became.

The assistant also listed estimates for managers, who cannot open the Estimates page. It now answers about estimates only for admins and owners, like the rest of the app.

Fix

AI clients that ask for every permission can now connect

Some MCP clients, Gemini CLI among them, ask for every permission Time Trakkr lists when they connect. The list included one used only by the Time Trakkr command-line tool for coding agents, which cannot be combined with any other, so their sign-in stopped at "Invalid authorization request". That permission is no longer listed, and those clients connect with read or read/write access as you choose. Nothing changes for clients that already connect.

Fix

Configure GitHub App brings you back to Time Trakkr

Configure GitHub App, under Settings → Agent time → Repositories and on a project's page, sent you to GitHub with no way back. GitHub saved your change, then showed a bare error from Time Trakkr. It now returns you to the page you started from, with the repository list fetched again.

If you install or change the App from GitHub itself, for example from your organization's settings, GitHub still saves the change, and Time Trakkr now shows a short page saying so, with a link back. That return does not connect anything to a workspace: Dependency Watch is still connected from its own page.

Improved

The privacy policy now covers agent time and the contact form

The Privacy Policy now lists what is reported for a coding-agent session in a mapped repository: the session's times, the repository's address and branch, the model and token counts. It also says what is never sent: prompts, the agent's replies, code, file contents or diffs. It describes what an agent in the cloud sends when it checks in, and the operator contact kept for an agent-operated workspace.

It describes the contact form instead of the invite waitlist, which no longer exists, and names OSV.dev, which receives package names and versions for Dependency Watch. It also says that Microsoft is used for Outlook calendar only, and that file storage holds profile photos and expense receipts as well as logos and employee documents.

Nothing about what Time Trakkr collects has changed. The policy now describes it more completely.

Fix

Slack's /track says truthfully whether time is billable

The /track help used to say you could add non-billable to log time that isn't billable. That never worked: the entry followed its task, as every entry does, while the reply in Slack still said "non-billable". So time could be recorded as billable while Slack told you it wasn't.

The reply now reports what was actually stored. Time on a non-billable task says so; if you type non-billable against a billable task, the reply says the time is billable and why. To log time that isn't billable, use a task that is marked non-billable, or ask an admin to change the task.

Fix

Daily rollups skip the weekend, and the overdue-invoice notice says who gets it

A project rollup set to Every weekday morning was also arriving on Saturday and Sunday. It now goes out Monday to Friday only, in your workspace's timezone. Monday's covers Friday, Saturday and Sunday together, so Friday's work is not dropped now that Saturday sends nothing. Weekly rollups are unchanged.

Under your notification settings, the overdue-invoice notification used to read "An invoice I raised goes overdue". It goes to every owner and admin, whoever raised the invoice, so it now reads "An invoice goes overdue (owners and admins)". It appears on the bell only. It has never sent an email, and it still doesn't.

New

Coding agents in the cloud can check in their time

A coding agent running in a vendor's cloud has no plugin beside it, so until now none of its time was recorded. It can now check in instead: it tells Time Trakkr when it starts and finishes work in a mapped repository, through the Time Trakkr connector or the API, and Time Trakkr's server counts the time between those calls.

The agent never sends a duration or a time, and a check-in that tries is refused. A silence of more than 15 minutes counts nothing, a turn counts for at most 4 hours, and an agent that forgets to say it finished is counted only up to its last check-in.

Checked-in time is never accepted automatically. It waits in your pending list, labeled Checked in by the agent, until you accept or discard it, and is billed at your workspace's agent rates once accepted. A repository nobody has mapped is not tracked, and nothing about it is kept.

It does not record agents that run on a credential many people share, such as the GitHub Copilot cloud agent or Claude in Slack: add those as agent members. See agent time for agents in the cloud.

Improved

Agent time tokens for a cloud environment can expire

When you create a personal access token with Agent time only ticked, you can now say it is for a cloud environment, such as a coding agent's secret store in a vendor's cloud. A token like that is offered an expiry, 90 days unless you pick 30 days, a year or never, because it lives somewhere you do not control. Your token list shows when each token expires.

An expired token is refused with a message saying when it expired, so an agent or script knows to ask for a new one. Tokens you already have do not expire, and nothing changes for them.

New

Ask the assistant to send a form

An HR admin can now ask the assistant, or Claude through the connector when it is signed in as you (a workspace API key is read-only and cannot send), to send a form: "send the Self-Assessment to the design team, with a manager evaluation". It takes the same choices as Send a form under HR → Forms.

It always previews first: the name the form will get, how many people it goes to, every form it would create and who writes each one, and anyone with nobody to write their manager evaluation. Nothing is sent until you say so.

It will not send without that confirmation, it is refused for anyone who is not an HR admin, and it does not choose colleagues for peer feedback — that stays on the Sent tab.

Improved

The Detailed report says how agent time was measured

Beside the Source column, the Detailed report and its CSV export now have an Evidence column. For agent time it says how the time was established, such as Measured, by the plugin counting the agent's turns; for a person's own time it is empty. The agent work report, and the assistant's answers about it, split agent hours the same way.

Agent time accepted before this change carries its evidence too, copied from the session it was accepted from. The column is empty only where that session was deleted first. Nothing about the hours or what is billed changes.

Improved

Your tokens show which agent last used them

Your list of personal access tokens now shows which coding agent last used each token, for example from Cursor, beside when it was last used. A token used by an agent you do not run means somebody else has it, perhaps through a secret your team shares, and you should revoke it.

It records the agent's name and nothing else: not where the request came from, and nothing about the work. It never slows down or refuses a request.

Improved

Agent sign-up names an operator and accepts the terms

An agent signing up through POST /api/v1/agent_signup now sends two more fields. operator is the name and email address of the person or company responsible for the agent, who is the customer under the terms. accept_terms is the version of the "Agent-operated accounts" terms it accepts on their behalf, currently 2026-11-06.

A sign-up without them, or naming an old version, is refused with a message that gives the current version and where to read it, and the agent's challenge stays usable. The acceptance is recorded with the workspace: the version, when, and the network it came from.

The operator's email is kept only for notices about the workspace, such as a suspension or a change to the terms. It is not a login, it is not a member of the workspace, and nothing is emailed to it yet. It is deleted with the workspace.

Fix

Agent sign-up returns links an agent can use

Agent sign-up and recovery tell the agent where to go next: the API base, the MCP server and, when sign-up is refused for the terms, where to read them. Those links named an internal server address instead of the one the agent called, so an agent following them reached nothing. They now use the address the request came in on.

Improved

Terms for agent-operated workspaces

The terms of service have a new section, "Agent-operated accounts", for workspaces created by an agent through agent sign-up. It says that the person or company responsible for the agent (the operator) is the customer and answers for everything the agent does, what an agent-operated workspace may not be used for, and that we may suspend or delete one without notice for abuse.

It also sets out idle expiry: a workspace none of whose tokens is used for 30 days is suspended, and is deleted, with all of its data, 30 days after that unless its recovery key is used to restore it. Deleted data cannot be recovered. Agent-operated workspaces come with no uptime or support commitment.

The section has a version, 2026-11-06, so it is always clear which wording an agent accepted. Nothing changes for workspaces run by people.

New

Agent time now records Cursor

The Time Trakkr plugin now records Cursor's agent, in the editor's agent chat and in the Cursor CLI, cursor-agent, interactive and with -p, in any repository an admin has mapped to a project. Restart the editor after installing, so it loads the plugin. Install it once: cursor-agent plugin marketplace add https://github.com/Time-Trakkr/agent-plugins, then /plugins in cursor-agent, Marketplace, timetrakkr, Install for you. Then run npx @timetrakkr/cli cursor setup once: it shows the two entries it adds to ~/.cursor/hooks.json and asks first, and with them whole turns are counted.

What it will not do:

  • Count whole turns without cursor setup. Without it, the Cursor CLI tells the plugin when the agent uses a tool, but not when you send a prompt or when the agent stops, so a turn counts from its first tool call to its last. That is lower than the time the agent worked, never higher.
  • Record Cursor cloud agents. The plugin is not on that machine.
  • Send token counts. Token cost for Cursor sessions shows "not captured".
  • Leave out permission waits. Cursor does not report them, so they are counted.

See supported agents and setup.

Improved

Idle agent-operated workspaces are deleted 30 days after suspension

An agent-operated workspace none of whose tokens is used for 30 days is suspended, and its recovery key restores it. If it is still suspended 30 days later, the workspace is now deleted, with everything in it: its members and tokens, time, projects, invoices, the operator contact and the agent identity if it owns no other workspace. Deleted data cannot be recovered.

Using the recovery key at any point before then lifts the suspension, and the workspace is not deleted. A workspace switched off for any other reason, such as by us for abuse, is never deleted automatically, and workspaces run by people are never touched.

No warning is emailed first: agent-operated workspaces send no email, and there is no way yet to send a notice to the operator.

Fix

Small fixes from the past month

  • A form that cannot reach Time Trakkr says so. If something between your browser and Time Trakkr answered instead of the app, a form used to show a technical error and save nothing. It now says Time Trakkr could not be reached, that nothing was saved, and that what you entered is still there to try again. A questionnaire also keeps your answers on this device until you submit it, so a reload does not lose them.
  • Help and your account stay in view. In a short browser window, the sidebar's Help link and account menu could be pushed off the bottom. They now stay on screen, and the links above them scroll instead.
  • Peer feedback is in name order. HR now sees the peer feedback on a review listed by the name of the person who gave it, in an order that no longer changes from one load to the next.
  • Product tips go only to people who joined after they were switched on. The occasional email with a tip about Time Trakkr is now sent only to people who joined a workspace after tips were switched on, so people who were already using it are not sent a series meant for newcomers.
Fix

A tab left open across an update reloads instead of failing

If you left Time Trakkr open in a tab while a new version went out, the next link you clicked could fail with an error instead of opening the page. The old page was trying to run parts of the new version it could not understand.

Now the next click after an update loads the page in full, on the new version.

Tabs opened before this fix went out may see the old behavior once more.

New

Agent time: record what your coding agents work on, and bill it at its own rate

Agent time is new, and free on every plan, for teams that use coding agents. It records how long an agent worked in one of your repositories, for the person who ran it, and turns that into time entries on the right project, priced at an agent rate rather than the person's own.

Setting it up. An admin switches it on under Settings → Add-ons, which opens a short checklist: set a default agent rate, and connect the repositories you want tracked, each one mapped to a project. Each developer then installs the plugin for Claude Code or Codex with one command, and signs in through the browser the first time a session starts.

What is counted. The time from a prompt being sent to the agent stopping, turn by turn. Time between turns, and time spent waiting for you to approve something, is not counted. Only the repository, branch, model, token counts and timings are recorded; prompts, file contents and diffs never are.

What happens to it. Agent time waits in a pending list on the person's timesheet, where they can accept it, move it to another project or task, or discard it. At the end of each day, agent time from a connected repository is accepted automatically, unless an admin turns that off. Time mapped by a .timetrakkr.json file in the repository is never accepted automatically. People can still reduce or delete an accepted entry, and they cannot make it longer than the agent measured.

How it is billed. Accepted agent time counts toward project budgets, profitability and invoices, on invoice lines of its own labeled "(agent time)". It is kept out of the person's own hours, utilization, reminders and weekly approval.

What it does not do: it does not record an agent in a repository that is not mapped to a project, and it records nothing until the workspace has a default agent rate.

New

The agent time plugins install with one command

The agent time plugins are published at github.com/Time-Trakkr/agent-plugins, and the command-line tool they share is on npm as @timetrakkr/cli. Nobody has to load a plugin from a local copy any more.

For Claude Code:

claude plugin marketplace add Time-Trakkr/agent-plugins && claude plugin install timetrakkr@timetrakkr

For Codex, codex plugin marketplace add Time-Trakkr/agent-plugins && codex plugin add timetrakkr@timetrakkr, then trust the plugin's hooks with /hooks.

The first session in a connected repository, on a computer that is not signed in, opens the browser: pick the workspace and press Allow access. The sign-in can report agent time and read the repository rules that apply to you, and nothing else.

At release, agents that run in the cloud were not supported; later entries say what has changed since. See supported agents and setup.

Improved

Every integration on one page, each with its own guide

The website has an integrations directory at timetrakkr.com/integrations/all: calendars, Slack, Jira, Claude and other AI clients over MCP, scheduled backups, imports from Harvest, Toggl and Clockify, the REST API and webhooks, and Google sign-in, filtered by category.

Each integration has its own page with the setup steps from its help article, the plan it needs, common questions, and a list of what it will not do. That last list is there on purpose: an integration that quietly does less than you assumed is worse than one that says so up front.

Nothing changes inside your workspace. You still connect integrations from the Integrations page in the app.

Fix

Negative amounts put the minus sign first

Negative amounts of money, such as the profit on a project that costs more than it earns, showed the minus sign after the currency symbol: $-10,891.50. The sign now comes first, -$10,891.50, and the amount is still shown in red where it was before. An amount that rounds to nothing shows as $0.00, not $-0.00.

Fix

Projects show what has been invoiced for them

On a project's page, Invoiced read $0.00 and the Invoices tab was empty for any invoice generated from tracked time and expenses. They only counted lines typed against the project by hand. Both now count everything billed for the project, including invoices you have already sent, so your existing totals appear without anything to redo.

When one invoice line covers more than one project, such as an expenses line that groups one category across a client's projects, each project counts its own share of that line rather than the whole of it. A written-off invoice still counts as invoiced. A deleted invoice or line does not.

Improved

Set a project's billing type and fixed fee through the API and the assistant

A project's billing type and fixed fee could only be set on its Edit page. They can now also be set with billing_type and fixed_fee on the REST API's project create and update, or by asking the assistant. Project responses include both, for managers.

A fee is only accepted for a fixed-fee project. Changing a fixed-fee project to another type keeps its fee, as the Edit page does. On a fixed-fee project, reports count the fee spread over the billable hours logged to it.

Fix

The budget report API says which kind of budget a project has

budget_by on GET /v1/reports/project_budget read project_hours on every row, including projects with only a fee budget, only per-person budgets, or only an internal cost budget. It now says what budget actually is.

Projects with an hours budget still read project_hours, whatever else they have, so an integration that checks for it keeps working. A project without one has budget null, and budget_by names the budget it is listed for: project_fees, person_hours or internal_cost, in that order when it has several, or none. The figures themselves are where they were: the fee_budget fields, budget_people and cost_budget.

Improved

Claude knows which Time Trakkr tools only read and which change things

Every tool Time Trakkr offers Claude through the connector now carries a plain name, such as "Log time" or "Draft an invoice", and says what kind of action it is. Tools that only read, such as looking up hours or budgets, are marked read-only. Tools that change or remove something that already exists, such as setting a budget or approving a timesheet, are marked as changes, so Claude can ask you before running them.

This does not change what any tool can do, or who may use it. A read-only API key still sees only the read-only tools.

When a tool call fails, the message now names the field that was wrong, and an internal database error is no longer passed back word for word.

Fix

Switching workspace from the account menu works on custom domains

If you belong to more than one workspace and one of them has its own domain, choosing another workspace under Switch account in the account menu did nothing: the page stayed on the workspace the address belonged to. It now takes you to the other workspace's own address and keeps you signed in on the way.

New

Add agents to your team as members

Under Settings → Agent members, owners and admins can add an agent — a cloud agent, a CI job, a third-party agent on MCP — as a member with its own name, role and rates. It is given a token, shown once, and logs its time through the API or MCP. That time shows on the Team list, timesheets, reports and invoices like anyone else's, labeled as an agent, and it goes through the same approval as everybody else's week.

Agent members are free and are never a paid seat. They have no email and no sign-in: they are never sent reminders or notices, and they cannot open the web app. They are left out of HR records, leave and reviews.

Their time is held to limits a person's is not: every entry needs a start and an end, and an entry that overlaps another, ends in the future, or takes the day past 24 hours is refused. Rotate a token at any time; disabling an agent member stops its token at once.

This is separate from agent time, which records a coding agent's work as time belonging to the person who ran it.

New

Agents can run a workspace of their own

An agent solves a short proof-of-work challenge and creates an account and an agent-operated workspace in one API call, getting an owner token and a recovery key. From there it adds other agents as members, tracks time, approves or returns their weeks, and invoices clients — through the REST API and MCP. The agent quickstart in Help, also at /llms.txt, walks through it.

An agent-operated workspace sends no email at all. Invoices are delivered as a share link, can only be issued once the workspace has declared a seller name and address, and say they were issued by an agent-operated account. A workspace whose tokens go unused for 30 days is suspended, and its recovery key restores it.

Fix

The budget report names clients, scales monthly budgets and lists everyone's hours

Clients. Each row in Reports → Budget vs actual, and in the /reports/project_budget API (client_id, client_name), now says which client the project belongs to.

Monthly budgets over a longer range. A 20 hour monthly retainer viewed over a quarter is now measured against 60 hours, not 20, so a project on plan no longer reads as 40 hours over. The API keeps budget as one month's allowance and adds months_in_range, budget_for_range and budget_remaining_for_range; the assistant does the scaling itself.

Everyone's hours on per-person budgets. On a project where people have their own budgets, Budget by person now also lists whoever logged time without one — assigned or not — marked No budget. A project budgeted only per person now appears on the report at all. Projects without per-person budgets do not list everyone.

Budget settings. Removing a project's budgets while setting its alert percentage in the same request is refused with a message that says so, rather than claiming the project "has no budget yet". A project with no budget now reports no alert percentage instead of 80%. Showing progress to everyone is not itself a budget, so an alert percentage sent to a project with only that is refused too, rather than accepted and then shown as empty.

Alerts that are off stay off. Unticking Send email alerts on a project now switches the alerts off; it used to save 100%, so the email still arrived once the project went over budget. A budget with alerts off reports no alert percentage over the API and the assistant, rather than 80%, and changing its hours or other settings no longer switches the alerts back on.

New

Group a client's projects into engagements

If you open a project for each phase of a piece of work, or for each year of a support contract, you can now group them into an engagement and see them together.

  • Create engagements on a client's page, or on a project's Edit page while choosing one.
  • Group or filter the report builder by Engagement. This carries through saved, shared and published reports.
  • See Budget by engagement in the Time report: each engagement's hours budget, and for managers its fee budget, against what was logged, with its projects and people beneath it. Archived projects count, so a client's support hours can be followed across years.

Each engagement belongs to one client. Two clients can each have one called "Support", and a project can only join its own client's engagements. Deleting an engagement keeps its projects. Engagements can also be managed through the REST API and the assistant.

New

Fee budgets measure what the work is worth

Set a Fee budget on a project's Edit page, beside the hours and cost budgets. Time Trakkr values every billable hour logged on the project at its hourly rate and measures that against the fee budget: on the project page, in Reports → Budget vs actual (a monthly fee budget is scaled to the range you pick), in the /reports/project_budget API, and through the assistant, which also includes fee budgets when you ask which projects are near their budget.

It has its own resets every month option and its own alert percentage; the alert email goes to owners, admins and project managers who can see money. Fee budgets imported from Harvest ("total project fees") now show too.

On a fixed-fee project the fee budget still uses the hours at their rates, not the fee, so it tells you when the work has outrun what it was meant to be worth. It is money, so members never see it, even on a project that shows its budget to everyone. Expenses do not count against it.

Members also no longer see a project's hours on the Budget vs actual report unless that project shows its budget to everyone, which is the rule the project page already followed.

Fix

Fixed-fee projects report their fee as revenue

Reports used to price a fixed-fee project's hours at hourly rates, as if it billed by the hour, so its revenue and profit said nothing about what the client actually pays. Now each billable hour earns an equal share of the fee: a project with a fee of 10,000 and 40 hours logged earns 250 an hour, a month with 10 of those hours shows 2,500, and the people, tasks and months add up to the fee.

Because the share is spread over every hour logged so far, a past month's figure goes down as more hours are logged. The project as a whole still earns exactly its fee.

A fixed-fee project with no fee set now earns 0 and is marked no fee set, on reports and on the project page. Enter its fee on the project's Edit page.

Profitability, the report builder (including saved reports), the REST API's time reports and the assistant all count revenue this way, so they agree with each other. Cost is unchanged. Invoicing is not: uninvoiced amounts and drafted invoices still price fixed-fee time at hourly rates.

Fix

Harvest imports bring fixed fees and budgets

Importing from Harvest used to drop fixed fees and budgets. A fixed-fee project arrived with no fee, and a billable project with no hourly rate was marked fixed-fee when Harvest bills it at nothing. Budgets did not come across at all.

Running Integrations → Migrate from Harvest → Import clients, projects, tasks & people again now fills these in:

  • Fixed-fee projects get their fee, and billable projects without an hourly rate become hourly at a rate of 0.
  • Hours-per-project budgets, total-project-fee budgets and per-person budgets are imported, including monthly resets and the over-budget alert percentage. Fee budgets show to managers on the project page and the budget report (see the fee budget entry).

It will not replace anything you set here. A budget or a person's budgeted hours that already exist are kept, and so is a fee or billing type you changed since the last import. Per-task budgets are still not imported. The import summary counts the projects that had one, so you know which to budget by hand.

Fix

A reused external reference is refused with a reason

Each external_reference.id can be used once per workspace — that is what lets an import run twice without duplicating time. Sending one that is already taken, for example by an earlier import, used to fail with a server error and no explanation. It now returns 409 with a message, and nothing is saved. Look up the entry that holds it with GET /api/v1/time_entries?external_reference_id=…. A deleted entry keeps its reference, so a re-import cannot bring it back.

Fix

Projects with the same name are no longer merged in reports

Reports grouped projects, clients and people by name. Two projects called "Support" for different clients appeared as one row holding the hours and money of both. This happened in the report builder, profitability, saved, shared, published and scheduled reports, comparisons, the assistant and the REST API.

Each is now its own row. Where a name is shared, the label says which one it is: a project shows its client, as in "Support (Northwind)", plus a number when two projects with that name share a client name, and a client or teammate gets a number, as in "Alex Kim (2)". Names nobody else uses are unchanged. Time reports over the REST API also include project_id, client_id, client_name and user_id, so integrations can match rows without relying on names.

Tasks still combine by name across projects, so "Design" covers all design work. Add the project as a field to see it per project.

REST list responses also had broken pagination links. Behind our proxy, the links in every list pointed at an internal address. They now use the address you called. Time reports over the API are now paginated as their total_entries and links always said: each page holds per_page rows (100 by default, up to 2,000). Before, every page returned every row, so an integration that read only the first page of a large report now needs to follow next_page.

Fix

A clear "reconnect your calendar" message when drafting stops working

If you revoked Time Trakkr's access to your calendar, changed your password, or the connection expired, Draft from calendar used to fail with a block of technical text from Google. It now tells you that the calendar stopped accepting Time Trakkr's access, and to disconnect it under Integrations and connect it again.

New

Change several entries at once from the Detailed report

Every row you could change with its pencil now has a checkbox. Tick the rows you want, or tick them all from the header, and a bar above the table offers three changes: Change project and task, Change date and Delete. Delete asks you to confirm first.

Each entry is changed by the same rules as editing it alone. Approved and invoiced entries can't be ticked, a submitted week can't be moved into or out of, and whatever couldn't be changed is listed with the reason while the rest go through. Up to 100 entries change at a time.

There is no "mark billable" option: an entry is billable when its task is, so moving entries to a billable or non-billable task is how that changes.

New

Reopen a week you already approved

On Approvals, switch to Status: Approved and select Reopen on a person's row. Their entries for that week unlock and go back to unsubmitted, they are told it was reopened, and the reopen is recorded in the audit log.

It will not reopen a week with any time on an invoice: the whole week stays approved and you are told how many entries are billed. Deleting the draft invoice releases them. Members who approve time as a project manager cannot take an approval back, and a manager cannot reopen their own week while manager self-approval is off.

October 2026

Improved

Budget alerts say which client the project is for

Budget notifications used to name only the project. When several clients each have a project called something like "Development Monthly Retainer", two alerts on the same day read identically apart from the percentage.

They now read "Development Monthly Retainer for Northwind is over budget — 112% used". Alerts already in your notifications show the client too, and if a client is renamed, its alerts show the new name.

Budget alert emails already named the client; the notifications now match them.

Fix

A double-clicked Start no longer leaves a 0:00 copy of your entry

Starting a timer while another is running replaces it. A replaced entry that held no time and nothing typed was already discarded, but one with a note was kept, so double-clicking Start after typing a note left a 0:00 copy of the same entry beside the real one.

A replaced entry is now also discarded when it holds no time and is identical to the one you just started: same project, task, note and Jira issue. Starting different work, or the same work with a different note, still keeps the first entry, even at 0:00.

Fix

Morning calendar drafts now remember how you classify your meetings

When you confirm a calendar draft, Time Trakkr remembers the project and task you chose, and the next instance of that meeting (the same recurring series, or the same title) arrives already filled in. The Draft from calendar button has always done this, but the automatic 7am drafts did not: they asked the AI to guess every meeting afresh each morning, which is why recurring meetings kept arriving with the wrong client, project or task.

Both now work the same way. Meetings you have never confirmed are still suggested by the AI.

It remembers what you choose on the draft when you confirm it. Drafts that were already waiting on your timesheet before this change keep the suggestions they arrived with. Set the right project and task on those as you confirm them, and from then on the mornings will follow your choice.

New

Two weeks' notice before complimentary access ends

When a workspace has People, Schedule or another paid feature at no charge, its owners and admins now get one email two weeks before that ends. It names everything that is ending and the date, and links to Settings → Billing.

The email also says what does not change: nothing you have recorded is hidden. HR records, bookings and submitted weeks stay readable and exportable, and only adding new ones needs the paid plan.

If the end date is moved, the new date gets its own notice. Members who are not owners or admins are not emailed.

Fix

Billing shows when a canceled plan ends

Canceling the paid plan takes effect at the end of the month already paid for. Until now, Settings → Billing kept saying "subscription active" for that whole time, as if it would renew. It now says "Paid until November 28, then Free," marks the Paid plan with its end date, and points to Manage billing in case you change your mind.

Ask the assistant whether you are still on the paid plan and it gives the same date. Nothing about canceling itself has changed: it still happens in Stripe's billing portal, and you keep the paid features until the date shown.

Fix

Budget reports count the same hours as the project page

Settings → Preferences → Project budget counting decides whether non-billable time uses up a project's budget. The project page always followed it, but the Budget vs actual report, a person's budget on a project, the /reports/project_budget API, the weekly project rollup and the assistant counted every hour regardless. A project could be under budget on its own page and over budget everywhere else.

They all follow the setting now, and the assistant says which hours it counted. A task's budget still counts every hour on that task.

New

Preview a form or survey before you send it

Preview now sits beside Send form and Send survey, and on each template's own page. It shows the questions as the people answering will see them: help text, choices, required marks and the anonymous or named notice. When a form goes out with a manager evaluation or peer feedback, each has its own tab.

You can click and type in the preview to see how the form feels to answer. Nothing you enter there is saved, and opening it sends nothing.

Fix

Changing a task to non-billable can now update the time already logged to it

An entry's billable setting is decided when the time is logged. Switching a task to non-billable later left its earlier entries billable, so a project page could show Non-billable 0:00 above a table full of non-billable tasks, and a budget counting only billable hours counted them anyway.

A project's Edit page now lists, under Tasks, any task whose earlier entries disagree with it, with the number of entries and hours. Managers can bring them in line in one click, or leave them as recorded. Nothing changes unless you choose it.

Approved and invoiced entries never change: they keep the setting they were approved or billed with. On a non-billable project nothing becomes billable, whatever the task says. The assistant can find these tasks and apply the change too.

Improved

Terms and privacy policy updated for the paid plan

The Terms of Service no longer describe an invite-only beta. They set out the free plan and the paid plan, how members are counted and charged, what happens when you cancel, and that cancelling or a failed payment never hides or deletes anything you have recorded. Workspaces with complimentary access keep it, and cancelling a subscription does not remove it.

The Privacy Policy now lists the billing details kept for a subscribing workspace, and adds Stripe to the providers we use. Card details go only to Stripe; Time Trakkr never receives them.

New

Time Trakkr has a paid plan, and anyone can start free

Time Trakkr is out of its invite-only beta. Anyone can sign up and start a workspace, free.

Free stays free, for as many people as you have: time tracking, invoicing and estimates, expenses, profitability and margin, every report, export and import, the API, and agents over MCP.

One paid plan, $10 per member per month, adds People (profiles, documents, time off and reviews), Schedule, timesheet approvals, scheduled backups, email to clients from your own domain, and reports delivered on a schedule. There are no tiers above it. The workspace owner subscribes under Settings → Billing, and pays through Stripe.

If your workspace was here during the beta, the paid features you were already using stay free, with no end date. Settings → Billing shows what that covers.

Nothing you have recorded is ever hidden behind the plan. If a subscription ends, HR records, bookings and submitted weeks stay readable and exportable; only adding new ones needs the paid plan.

Improved

Billing has its own page in Settings

Your plan used to sit at the bottom of the Branding page. It now has its own page, Settings → Billing, which shows the plan, how many members a subscription would count, and any complimentary access with the date it ends.

Nothing is charged yet. When the paid plan starts, workspaces get notice first, and the workspace owner will subscribe from this page through Stripe.

New

Send client email from your own domain

Email to your clients can now come from your own domain, such as [email protected], instead of Time Trakkr's address. Add your domain under Settings → Client email, publish the DNS records it shows, and check verification. Replies go to that address.

Reminders and other mail to your own team still come from Time Trakkr. If the domain stops verifying, or the plan ends, client email goes from Time Trakkr again, so nothing ever fails to send.

Fix

Long written answers in reviews and surveys now save

A written answer in a review or a survey used to stop saving at 5,000 characters, about 800 words. Anything past that was refused each time the field saved, so a long answer looked fine on screen and was gone after a reload.

Answers can now be four times as long. The box stops accepting text at the limit, and a count appears as you approach it, so nothing you type can be refused.

Improved

A sending limit for email to people outside your workspace

Time Trakkr now limits how much email each workspace sends to people outside it: up to 60 recipients an hour and 300 a day, across invoices, estimates, shared reports and invitations. Mail to your own team doesn't count.

Few workspaces will ever notice. If you reach the limit, nothing is sent and the message says when you can try again. An invitation or shared report is still created, and its link is shown, so you can send it yourself. Contact support if you need a higher limit.

Improved

Reconnect a backup provider without starting over

A backup whose provider stopped accepting Time Trakkr's access, shown as an error like "Google Drive token refresh failed (401)", used to offer only Remove. Fixing it meant deleting the destination and adding it again, losing its run history.

Now the card says what happened and offers Reconnect. You sign in to the provider again and come back to the same page. Every destination on that provider uses the new access from its next run, and its schedule, retention and history stay as they were.

Reconnecting does not fix an outage on the provider's side. Those errors clear on their own at the next run.

New

Send a client their hours report every week or month

Under Reports → Saved reports, Schedule now sends a summary report to a client on its own, every week or every month. Each delivery is a new snapshot of the period just ended, emailed at 7am your time to the addresses you choose. It shows the same hours-only view as sharing by email, and the reader can be asked to confirm their address first.

Schedules can be paused, resumed or deleted at any time, and the list shows when each one last ran and what went wrong if it did not. It will not catch up on missed periods, and it will not send a detailed report.

Scheduled delivery is part of the paid plan. Sharing a report by hand stays free.

Fix

Joining by company email domain now needs a confirmed address

If your workspace lets everyone at your company domain join, a password sign-up used to join immediately. Signing up only proves someone typed the address, so anyone could have registered a colleague's address and joined.

Now they are emailed a link and join when they open it. People who sign in with Google join straight away as before, because Google has already confirmed the address. Nobody who has already joined is affected.

Until then, the page they land on says which workspace they will join and lets them send the email again, whether they signed up on the main site or at the workspace's own address.

Fix

The assistant finishes long requests, and says so when it cannot

When you asked the assistant to do a lot in one go, such as backfilling a few weeks of time entries, its reply could run out of room partway through. It then showed you its opening line ("Let me log all 28 entries at once!") as if that were the answer, and nothing was logged. It now has the room to do that much at once. If a reply is ever cut off, it tries again in smaller steps instead of stopping silently.

When a request needs more steps than one message allows, the assistant now says how many actions it completed before stopping. Say "continue" and it picks up from there, rather than starting over.

The assistant also knows today's day of the week, so "every weekday for the last five weeks" lands on weekdays.

Thumbs up and down now appear on the newest answer as soon as it arrives, instead of only after you reload the page.

Improved

Delivery team roles in the standard set

You can give each person on a project a role under Edit project → Team. The standard set of roles you can add under Team → Assignments now covers a software delivery team: Technical Project Manager, Lead Developer, Backend Developer and Frontend Developer join Project Manager, Engineer and QA Analyst.

If your workspace added the standard set before, choose Add standard set again. It adds only the roles you don't have yet and leaves your existing ones alone.

Fix

Buttons no longer overlap the text around them

A few buttons in Time Trakkr are links underneath, such as "Connect your GitHub account" in a project's Agent time repositories section. Where one of those sat on its own line, it covered part of the sentence above it and the line below. It now takes up its own space like any other button.

The Agent time repositories section on a project's page is also easier to read. "Manage all repositories" now looks like a link, and a line separates connecting GitHub from adding a repository by its address, so the two read as alternatives rather than one step.

New

See a project's scheduled hours against its reported hours, by role

Open a project and choose Team. Scheduled vs reported groups the project's team by the role each person holds on it — Engineer, Project Manager and so on — and shows each role's confirmed bookings and reported hours as two bars side by side, with the difference. Click a role to see its people.

It uses the dates already chosen on the project page. Scheduled hours are counted exactly as Schedule's Planned vs tracked counts them, and reported hours are every entry on the project, so the totals match both. Placeholder bookings for a role with nobody assigned appear under that role as Unassigned, and anyone who worked on the project without being on its team appears under No role rather than disappearing.

Scheduled hours are shown to managers, admins and owners, as on the Schedule page. Others who can see everybody's hours on the project see the reported hours by role. It shows hours only, never money, and it does not change any booking. The assistant can answer the same question for one project.

Fix

Picking a project on the Schedule shows each person's time on that project

Filtering the Schedule by a project showed the right people, but each with their whole allocation. Someone with 10 hours on the project and 30 elsewhere read as "40h of 40h" under that project's name.

Each person's row now starts with their hours on the project ("10h on Site Build"), and the bar and percentage measure that project against their capacity. Their whole schedule is still shown underneath, because it is what tells you whether they can take on more. The totals above the list add a "Scheduled on" figure for the project.

The assistant can answer the same question: ask how much of someone's month is on a given project.

New

Set budgets through the API and Claude

Every project in the REST API now shows its budget, and you can set it when you create or update the project. The fields are budget, budget_is_monthly, over_budget_notification_percentage, show_budget_to_all, cost_budget and cost_budget_include_expenses, named as in Harvest's API. A person's budget on a project is set on their user assignment, and a task's budget on the task.

The assistant can do the same: ask what a project's budget is set to, or tell it to change a project's, a person's or a task's budget. That includes Claude connected over MCP when you signed in to connect. An organization API key stays read-only.

Only managers, admins and owners can change budgets, as on the project's Edit page. Members see what they saw before: a project's hours budget only when it is shown to everyone, and never money.

One change for existing integrations: the project, user assignment and task endpoints now refuse fields they do not recognize with a 400, instead of ignoring them.

Improved

This month's hours on the project list

On the Projects list, Spent counts every hour a project has ever had, but Remaining on a budget that resets each month counts only this month. With nothing showing this month's hours, the two did not add up. A new This month column shows them for every project with a monthly budget. Projects on a fixed budget leave it empty, since Spent is already what their budget is measured against.

Also fixed: a budget that had run over showed its overspend with two minus signs and the wrong hour when durations are shown as hours and minutes. 8 hours 57 minutes over read "-9:-57"; it now reads "-8:57". This affected negative durations everywhere they appear, not just on the project list.

New

Add a project straight from its client

Starting a project for a client meant going to Projects, selecting + New project and finding the client again. Each client on the Clients page now has + Add project, and so does the Active projects panel on the client's own page. The New project window opens with that client already chosen.

The projects listed on a client's page now link to the projects themselves.

Owners, admins and managers can add projects, as on the Projects page.

New

Draft from your calendar every morning

Beside the Draft from calendar button on your timesheet there is now a checkbox: Draft automatically every morning at 7am. Tick it, and every morning at 7am in your own timezone that day's meetings appear as drafts on your timesheet, ready to confirm or dismiss.

It is off until you turn it on, and it only ever reads your own calendar. Nobody else can turn it on for you. It never logs time by itself: every draft still waits for your Confirm.

It drafts the day as it stands at 7am. A meeting added later that day needs the Draft from calendar button, and drafting a day twice never creates duplicates.

You can also ask the assistant to turn it on or off.

New

Questions that take more than one answer

When you add a question to a template, the type list now has Choose any beside Choose one (what used to be called Multiple choice). The person answering picks every option that applies: one, several or all of them.

A required Choose any question needs at least one option picked. There is no way yet to set a fewest or most to pick.

Survey results and peer feedback count how many people picked each option. Because each person can pick several, the counts can add up to more than the number who answered. Spreadsheet and PDF downloads list every option picked in one cell, separated by semicolons.

The assistant can add Choose any questions when it creates or edits a template.

New

Delete a project or client created by mistake

A project or client created by mistake, with no time, expenses, invoices or other history, can now be deleted for good. Archive it first. Then, in the archived view, owners and admins see Delete… beside Restore.

Anything with history can still only be archived. Trying to delete it says what is attached, for example "This project has 3 time entries, so it can only be archived." A client's projects have to be deleted before the client, and deleting cannot be undone.

The API reference now gives the status that DELETE requests actually return: 200 with an empty body, or 204 for removing a client contact. It previously said 204 for all of them. Deleting through the API still archives a project or client rather than deleting it.

Fix

Timesheet reminder emails now go out

If your workspace has timesheet reminders switched on, people should have been getting emails about time they had not logged, timesheets past their deadline and timesheets waiting for their approval. A fault stopped those emails from going out. They now arrive as your reminder settings describe.

Nothing that should have been sent before this fix will be sent late. Reminders start from the next one due.

Fix

Going to your workspace's address no longer asks you to sign in again

Typing your workspace's address, such as your company's own Time Trakkr domain, used to show the sign-in page even when you were already signed in. So did clicking the logo on that page. Your session was never lost, but it looked as if you had been logged out.

Now the address takes you straight to your timesheet when you're signed in. If you reach the sign-in page while signed in, it says which account you're using and offers Continue, or Sign in as someone else to switch accounts.

New

Client contacts over the API, and finding a client before creating one

For integrations such as Zapier that keep clients in step with another system:

  • Client contacts over the API. List, add, update and remove the people at a client at /v1/clients/{id}/contacts. An email can be on a client only once, ignoring capitalization. A second one is refused with a conflict that says which contact already has it. The Clients page now applies the same rule.
  • Look a client or project up by its whole name with name_exact. Projects can now also be filtered by a name fragment.
  • Find-or-create in one call with find_existing on creating a client or project. A match is returned as it is, unchanged.
  • Client responses include notes.
  • Archived clients and projects are left out of lists unless you ask for them with include_archived=true. Until now they were mixed in. If something you built relies on seeing archived records, add that parameter.
  • A project or contact can no longer be filed under another workspace's client. Both used to be accepted.

The assistant can now find clients that have no projects yet, read their notes and contacts, and, for owners and admins, add a contact or update one. It does not remove contacts.

Fix

The assistant explains why it could not answer

When a question failed because of the AI provider, such as a workspace's own Anthropic key being rejected, the assistant could show a long block of web page code where its answer belonged. The explanation was never shown.

It now says what went wrong and where to fix it, for example that an admin can update the key in Settings → Assistant. If the reason cannot be read at all, it asks you to try again in a minute instead of showing raw code.

This does not fix a rejected key. An admin still needs to update it.

Fix

Form downloads hold completed forms, and say when they will be ready

Downloading a form that nobody had filled in gave you a row for every person it was sent to, saying "Not submitted yet". That read as a list of forms, not a form. Downloads and printed copies now hold only the forms people have submitted, with their questions and answers, so you can review them offline.

While a form is still open, Download CSV and Print / PDF stay in place but grayed out. Hover over one, or tap it on a phone, and it tells you when it will be available: once you submit your form, or for HR, once someone has submitted theirs.

Drafts are still never exported.

Fix

The assistant is readable in dark mode

In dark mode every message in the assistant was drawn on the same light tan as in light mode. Each reply was a bright block on a dark page, and your own messages looked just like the assistant's.

Replies and your messages now have dark backgrounds of their own, different enough to tell apart. Links, tables and code inside a reply read clearly on them, and the selected conversation in the sidebar is a quiet highlight rather than a bright one. The note on a shared conversation no longer shows as a light strip, and a 👍 or 👎 you have chosen now shows as chosen. Light mode looks as it did.

New

Export forms and survey results as CSV or PDF

Forms and surveys can now leave the app:

  • A sent form, everyone's answers. On a sent form's page under HR → Forms → Sent, Download CSV gives one row per answer, with whose part it is, who wrote it, its status and the question. Print / PDF lays the same answers out person by person, ready to save as a PDF. Only what has been submitted is exported, never a draft.
  • One completed form. At the top of a completed form, Download CSV and Print / PDF give a copy of it. Whoever can open the form can export it, and nobody else.
  • Survey results. Beside the existing Download CSV, Print / PDF gives the results question by question. An anonymous survey's results are held back from the printed copy exactly as they are on the page, until it has closed with enough answers.

PDFs are saved from your browser's print dialog. HR can also ask the assistant for everyone's answers on a sent form.

Fix

A submitted form shows what you chose, and its email opens it

Four fixes to forms:

  • Your answers stay visible after submitting. A locked form used to draw every rating button alike, so the one you picked only showed when you hovered over it. The rating or choice you made is now filled in, and the others are faded.
  • Submitting takes you back to the top, where a line says it went through, instead of leaving you at the bottom of a long form.
  • The link in a form's email opens that form. It used to land on the Forms page and leave you to find it. If nothing in that form is yours any more (it was withdrawn, say), the page says so.
  • On a sent form's page, HR opens a person's answers with View answers, beside each part that has some, or by clicking the person's name. The status labels are no longer the only way in.
Improved

See who a form went to, and where each of them has got to

The Sent tab under HR → Forms is now a list. For each form it shows when it was sent, who it went to, how many of them are done (and how many are in progress or not started), and whether it is open. Two sends of the same template often share a name, and this tells them apart at a glance.

Click a form to open its own page. Every person it went to is listed with a status: Not started, In progress, or Done once every part about them is complete. Their self-assessment and manager evaluation sit beside it, along with who is writing the evaluation. Not finished narrows the list to the people who still owe something. Remind, Withdraw, Edit, Close and Delete are all on that page.

Ask the assistant about a sent form by name and it now answers for every form with that name, each with who it went to and how many are done.

Improved

Templates are a list, each with its own page

The Templates tab under HR → Forms is now a list of every template. Each row shows how many questions it has, what it has been sent as, and whether its questions can still change. Click a template to open it on its own page. Its questions, Rename, Duplicate and Remove are all there, along with Send a form with it.

Creating a template, or starting from a ready-made one, opens it straight away. A reload, or a link to a template, comes back to the same one.

Improved

Reviews written blind, survey thresholds, and reminders on for new workspaces

Reviews are written blind. When a form has both a self-assessment and a manager evaluation, whoever writes the evaluation sees the person's self-assessment only once they have submitted their own. Until then it says so in the person's file. An evaluation written after reading the person's account tends to answer it rather than evaluate them. HR admins who are not writing the evaluation see the self-assessment as soon as it is submitted, as before.

Each survey sets its threshold. When you send a survey you choose how many answers it needs before its results show: five unless you change it, and never fewer than three. Surveys already sent keep the three they were sent with.

Reminders start on in a new workspace, a week after sending, then weekly, and the day before a form is due. Workspaces that already exist keep the reminder settings they have, which are off unless HR switched them on, so nobody is suddenly chased about requests left open long ago. Switching one on now starts from a week rather than three days.

Improved

Clearer forms and surveys

Every form, survey and part of a form now shows its status the same way everywhere: Not started, In progress, Submitted, Shared or Acknowledged. A form you are filling in shows its name, whom it is about and when it is due.

Also in this change:

  • Forms and Surveys work on a phone: the list sits above the form instead of squeezing it.
  • Send a form starts from the template you are looking at.
  • A self-assessment counts as done once it is submitted. Nobody is asked to acknowledge their own answers any more; acknowledging is for an evaluation someone else wrote about you.
  • A template used by a survey can no longer have its questions changed under the answers already given.
  • The old Issue to everyone button on unsent drafts is gone. Delete a draft, and send the form to the people it is for.
Improved

HR has its own pages for forms, surveys and reminders

What HR sends now lives in one place, under HR in the sidebar:

  • Forms has three tabs: Sent, Send a form and Templates. Each sent form is a table, one row per person, with their self-assessment and manager evaluation side by side and a Remind button on the row.
  • Surveys has Sent and Send a survey. Once you send one, the page moves to Sent and opens it.
  • Reminders shows each kind of form as one line saying what it will do. Open a line to change it, and save them all with one button.

Old links and bookmarks to the previous addresses still work and take you to the new pages. Nothing about who can see or send what has changed.

New

An Inbox, and a tidier sidebar

Inbox lists everything waiting for you in one place, soonest due first: forms to fill in, a manager evaluation of you to read and acknowledge, surveys to answer, documents and profile details HR has asked for, and — if you approve timesheets — weeks waiting for you. The number beside it is how many. Each item opens where it is done.

The sidebar is now grouped by what you are doing:

  • Track — Timesheet, Expenses, Assistant and Reports.
  • Work — Clients, Projects, Tasks, and for managers Schedule and Approvals.
  • People — your People file and Time off. Forms and surveys are filled in from your Inbox, and completed forms are kept in your file; a line manager finds their direct reports' forms there under My team.
  • Finance — Invoices and Estimates, for owners and admins.
  • Manage — Integrations for everyone, plus Team, HR, Developer and Settings for the people who use them.

HR is new: one place for HR admins to send forms and surveys, manage templates and reminders, and set up time off.

Your personal access tokens and connected apps have moved from Developer to Integrations, beside your calendar, Jira and Claude connections. Developer now holds the organization's API keys and webhooks, for managers.

Every item in the sidebar has its own icon.

The assistant can tell you what is waiting for you, too.

New

Remind, rename, withdraw or delete a sent form

Under People → Forms → Set up and send forms, each sent form now has Edit, for its name and due date, and Delete…. Progress has Send a reminder beside each unfinished part, and Withdraw… beside each person. Send a reminder is also on the form in the person's People file.

A reminder goes by email and in the app straight away, at most once a day per person. One sent by hand counts toward the automatic reminders, so nobody gets two on the same day.

Deleting a form, or withdrawing it from somebody, works until the first submission. After that the answers are part of the person's People file, so the form can be closed but not deleted.

Also fixed: acknowledging a manager evaluation shared with you failed with an error. It works now.

New

Peer feedback on reviews

When you send a form with a manager evaluation, you can now add a Peer feedback template. After sending, choose who gives feedback on each person from their row on the Sent tab. Each colleague you ask finds it in their Inbox.

Feedback is only honest when it is safe to give, so who reads what is strict:

  • HR admins see each colleague's feedback with their name.
  • The manager sees it combined and without names: averages for ratings, and written answers in alphabetical order. They see it only once at least three colleagues have answered, and only after submitting their own evaluation.
  • The person sees the same combined view once their manager evaluation is shared with them.

What it will not do: let people nominate their own reviewers (HR chooses), or show one colleague's feedback on its own to anyone but HR.

Improved

A tidier People file, and filling in a long form

A person's file under People is now in tabs — Profile, Pay, Documents, Forms and, where it is switched on, Systems access — instead of one long page with the forms at the very bottom. The tab you are on stays put as you move from one person to the next, so HR can read everyone's forms in turn.

Filling in a form or a survey, a bar stays at the foot of the screen with how many questions you have answered, whether your last answer saved, and the Submit button. On a long performance review these used to be a long scroll away. A rating question says which end of the scale is low and which is high, unless its help text already does.

New

Totals at the top of the Schedule

Capacity now opens with the hours available to schedule (working time less leave), how many are scheduled, how many are still unassigned, and how much of the team's capacity that uses. Planned vs tracked opens with total planned, total tracked and the difference.

Both follow the dates and filters, so choosing a group gives that group's totals. Unassigned hours are added up person by person: one person being overbooked does not count as spare time for somebody else, and the hours over are shown on their own.

The assistant gives the same totals when you ask how much of the team's time is free, or whether you are on plan.

Fix

The API shows a member's token what the app shows that member

A personal access token, or a connected app, acts as the person who made it. For a plain member the API answered a few requests more generously than the app does for the same person. It now matches the app.

With a member's token:

  • Projects and tasks follow the workspace's "members only see assigned projects" setting, as the project picker does. A project the member is not assigned to is not listed, and asking for it by id answers 404, as do its tasks and assignments.
  • Other people are not listed. GET /users returns the member alone, and another person's id answers 404.
  • Rates on project assignments are blank.
  • Invoices and work roles answer 403, like the other manager-level endpoints.

Nothing changes for an organization key, or for a manager's, admin's or owner's token.

Logging or moving time now checks that the task belongs to the project, on every route, and answers 422 when it does not. Logging time to an archived project still works, as it does on the timesheet.

If an integration uses a member's token and relied on any of the above, give it an organization key or a manager's token.

Improved

Arrange the systems access list by dragging

Under Settings → Systems access, drag a system by its handle to change the order, or drop it onto another system to make it a sub-system — GitHub B onto GitHub A becomes GitHub A › GitHub B. A sub-system dragged to the end of the list goes back to the top level. The new order is the order on every profile.

Where a drop is not allowed — a system that has sub-systems of its own going under another, or two systems of the same name under one parent — the page shows it as blocked and says why.

Every row also has a Move… menu that does the same without dragging, from the keyboard or a phone. It replaces the Add under button on each row; to add new sub-systems, choose the parent in Add under below the list.

The assistant can move a system too: ask it to put one system under another, or back at the top level. Moving a system never changes who is marked as having access.

New

Edit review templates, and duplicate one for the next round

Under People → Reviews → Set up, every question now has Edit: change its wording, help text, type, choices, or whether it is required. Drag a question by its handle to reorder it, or use Move… on its row. Templates can be renamed and given a description.

Once a cycle using a template has been issued, its questions are fixed, because answers already given point at them. The page says which cycles use it, and Duplicate makes a copy — questions and all — that you can edit for the next round. The original stays exactly as the earlier cycles used it.

Templates written with their own numbering, such as "1. Technical skills", are now shown as written, without a second number in front.

The assistant can make the same changes for an HR admin.

New

Automatic reminders for forms people have not finished

Under People → My forms → Form reminders, HR admins choose, for each kind of form, when the first reminder goes, how often after that, and how many at most. There is also an option to send one the day before a due date. Reminders go by email and in the app, on weekdays during working hours, and stop the moment the form is done.

They are off until you switch them on. When you do, the next hourly check includes forms sent before that, so a long-open request will be chased too.

A questionnaire, review cycle or document request can have its own schedule, or none, chosen when it is sent. The page also lists the reminders sent recently.

New

Forms for named people, surveys for anonymous answers

There used to be two ways to send someone a set of questions, and both could be named. Now each has one job.

Forms (formerly Reviews) go to named people: one person, a few, a group or everyone. Each answers it about themselves, with an optional manager evaluation alongside. The completed form is kept in their People file. Use them for self-assessments, probation check-ins, anything where who said it matters. Send one under People → Forms → Set up and send forms.

Surveys (formerly Questionnaires) are always anonymous. Answers are stored without names, and results show only as tallies once a survey is closed and at least 3 people have answered. Surveys no longer have a named option: to ask named people, send a form.

Questionnaires already sent with names are still there, with their answers, under Surveys.

New

Completed reviews are kept in the People file

Open someone in People and the new Forms section lists every review sent to them, with their self-assessment and any manager evaluation of them. Read opens the answers.

The person sees their own forms on My profile. Line managers find their direct reports under People → Reviews → My team. HR admins see them on each person's record.

A manager evaluation is still private until it is shared: the person sees Not shared with you yet until then. Answers still being written are not shown to anyone but their author.

New

People are told when HR sends them something to fill in

When an HR admin issues a review cycle, sends a questionnaire or asks somebody for a document, each person it is for now gets a line under the bell and an email, with a button that goes straight to it. A reviewer with several reviews to write gets one message saying how many. A document pack is one message listing everything asked for, not one per document. A profile-details request already sent an email, and now adds the line in the app too.

The line in the app can be switched off per person under Notifications, like the others; the email is part of being asked, so it always goes.

Automatic reminders for anyone who has not finished are set up separately, under Form reminders.

New

Move a day's entries to another day, or copy a day with its time and notes

Below a day's entries, Move to another day… moves them to the day you pick, with their hours, notes, task and Jira issue, and takes you there. Untick any that should stay. Approved and invoiced entries cannot be moved; anything else that cannot move is listed with the reason.

On an empty day, the copy button is now two: Copy rows (no time or notes), as before, and Copy with time and notes, which copies every entry from your most recent day as it was. Copying never changes the earlier day, so to fix time on the wrong day, move it rather than copy it.

New

Send questionnaires, named or anonymous

Under People → My forms, HR admins can now send a questionnaire: a review template's questions sent to everyone, to one group, or to people you pick. Use it for a pulse survey, offsite feedback or a quick poll. People answer it under My forms and submit once.

Each questionnaire is either named or anonymous, and that is fixed when it is sent. Anonymous answers are stored without names or times. Their results show only once the questionnaire is closed and at least three people have answered, so nobody's answers can be singled out. Results show averages, counts per choice and the written answers, and download as CSV.

People are told when one is sent to them, in the app and by email. Reminders for anyone who has not answered are set up under Form reminders.

New

Send a review to one person, a few, or everyone

Under People → Reviews → Set up, Send a review replaces creating and issuing a cycle. Choose who it goes to — one person, people you tick, a group, or everyone — and the self-assessment template they fill in.

You can add a manager evaluation on its own template, so the manager answers questions written for them rather than the self-assessment's. It goes to each person's line manager, or to one person you choose. Before sending, the form says who has nobody to write theirs.

Sent reviews lists what has gone out, with progress for each. A sent review cannot be widened afterwards: to add somebody, send it to them.

Improved

Stricter checks when signing in and connecting apps

A few things about signing in and connected apps are tighter. Most people will not notice a difference.

  • After signing in you always land on a Time Trakkr page. A sign-in link that names a page we cannot be sure is ours now takes you to your timesheet.
  • Connecting an app follows your workspace's two-factor rule. If your workspace requires two-factor authentication and you have not set it up, the page for connecting an app now says so and points you to Account security, instead of letting you connect the app first.
  • An app gets what it asked for, or less. When you connect an app you can still untick "Log time, run timers, and draft invoices as you". The app cannot end up with more than the page showed you.
  • A connected app ends when its owner leaves. When somebody is deactivated or removed from a workspace, apps they connected to it stop for good. If they come back, they connect the app again.

If you connect an app and never finish, it is forgotten after a week and has to be set up again from the app.

New

Require a Jira issue on chosen tasks

On a project's edit page, tick Needs Jira beside a task, and new time on that task has to name a Jira issue — Engineering on one client can need one while Engineering on another does not.

The issue is checked against Jira when the entry is saved. An issue Jira does not have is refused; so is saving while Jira is not answering, with a message to try again, rather than letting an unchecked issue through. The rule holds everywhere time is logged: the timesheet, the timer (which asks when you start it), the assistant, Slack and the API.

Entries made before the rule are left alone: you can still fix their hours or notes. Only new entries, entries moved onto the task, and changes to an entry's issue are checked.

Improved

Fix an entry straight from the Detailed time report

When a report shows time logged against the wrong task, you can now fix it where you found it. The pencil at the end of a row opens the entry in the same dialog as the timesheet; save, and the row updates.

Owners, admins and managers can edit anybody's entries this way, and the change is recorded against them, the same as editing a timesheet. Everyone else can edit their own.

Approved and invoiced entries show a lock instead, because they have to be reopened before they can change.

Improved

Fix an invited person's email, and see them on People before they join

Correcting an address before somebody joins. On Team, click a person's name and change Work email. If you invited them at the wrong address, the old invitation stops working — otherwise whoever reads that mailbox could accept it — and Send invite to … sends one to the new address.

Only owners and admins can do this, and not on their own address, which is changed from your profile.

Invited people on People. Somebody invited and not yet joined is now listed under Invited, not joined yet, so you can set up their systems access and paperwork before their first day. They are not asked for missing details until they have joined.

New

Filter the schedule, see who is on a project, and what each booking adds up to

Filters. Under the dates on the Schedule page you can now narrow everything to one of your groups (a pod or team), to a team member, or to one project, picked from the same list as the booking form. Picking a project shows the people allocated to it, the way a group shows its people. Capacity and the bookings list follow every filter. Planned vs tracked follows the project only, because its figures are whole-project totals for everybody, and the page says so. Unassigned work is hidden while a filter is on, because it is booked against roles rather than people.

By project. The bookings list can be grouped by project instead of by person: who is allocated to each project, with the confirmed hours booked in the dates shown and tentative hours apart.

Booking totals. Every booking now shows what it adds up to, counting weekdays and skipping the person's public holidays — 176h for a month at 8h a day, not 248h. A booking that runs past the dates shown also says how much of it falls inside them.

The assistant can answer the same questions: ask who in a pod is booked on a project next month, and it will give each booking's total.

Improved

Systems access in bulk, and sub-systems such as client GitHub organizations

Add several systems at once. On Settings → Systems access, paste names one per line or separated by commas. Anything already on the list is skipped and named; the rest are added.

Tick one system for many people. People on a system's row lists everybody with a box each. Tick and clear as many as you need and save once. Only the boxes you changed are recorded.

Sub-systems. A system can have sub-systems one level down — a client's GitHub organization under GitHub, for example. Each is its own box, indented under its parent on profiles, and named "GitHub › Northwind" wherever it appears on its own.

Like the rest of systems access, these are records only: nothing is changed in GitHub or anywhere else.

Improved

A person's hours on the Team page open the entries behind them

On the Team page, a person's Hours for the week now open the Detailed time report for that person and that week, listing the date, project, task and notes of each entry. Their Billable figure opens the same view, showing only billable entries.

A figure of zero is not a link, because there is nothing behind it. Who can see whose entries has not changed: the report shows you only what it always did.

New

Agent time sets itself up in one page, and needs nothing from you day to day

Enabling Agent time under Settings → Add-ons now takes you to Settings → Agent time → Setup, a checklist that says what is done and what is left:

  • A default agent rate, which is required. You set it right there. Until a workspace has a rate for all its work, agent time records nothing, and each developer's agent says it is not set up yet, instead of collecting time that could never be accepted. A rate for one client, one project or one agent does not count as the default. Time from before the rate was set is not sent later.
  • Repositories: how many are connected, with a button to connect more.
  • Accept automatically, now on by default. Agent time from a connected repository becomes a time entry once the day has ended, without anybody accepting it. People can still reduce or delete such an entry, and deleting one puts its time back in their pending list. A workspace where an admin had turned it off keeps it off. Time mapped by a .timetrakkr.json file in a repository is still never accepted automatically.
  • Roll out to your team: the install command for each agent, and how to make the plugin required for everyone, once the plugins are published.

The plugins are not published yet. Until they are, the rollout step says so rather than showing a command that would not work, and developers load a plugin from a local copy as before.

The assistant can answer "is agent time set up?" too.

Fix

New tokens and keys copy with one click

When you create a personal access token, an API key or a webhook, its secret is shown once. It now has a Copy button beside it.

Selecting it by hand works better too: one click selects the whole secret, and nothing else. Dragging used to catch the Done button next to it, so a pasted token could end in "Done".

New

Pick agent time repositories from every GitHub account you can reach

Under Settings → Agent time → Repositories, press Connect your GitHub account. The list then holds every repository you can reach through the Time Trakkr GitHub App, across your own account and every organization it is installed on.

Configure GitHub App opens GitHub, where you can install the App on another organization or share more repositories with it. Press Refresh list when you come back. You can also paste a repository's URL into the search box to add a repository that is not listed.

The list is personal. It shows what your own GitHub account can reach, so another admin may see a different list. Rules already set keep working whatever the list shows.

Fix

Connecting GitHub now confirms the installation is yours

When you connect GitHub, GitHub now also asks you to approve Time Trakkr for your own GitHub account the first time. Time Trakkr uses that approval to confirm you can reach the installation you chose, before connecting it to your workspace. If your account cannot reach it, the installation is refused, with a message explaining why.

Connections you already have keep working. Nothing changes about which repositories are connected.

Fix

The notes field says when notes are required

If your workspace requires notes on time entries, the notes field in the time entry dialog and in the sidebar timer now reads Notes (required). It used to read "Notes (optional)" even when an entry without notes would be refused.

The rule itself has not changed. It is still switched on or off in your workspace settings, and it still applies wherever time is logged.

Improved

Each day on the timesheet shows its date

Each day in the timesheet's week view now shows its date under the weekday, so it reads "Wed Sep 2" rather than just "Wed". You no longer have to count along from the start of the week to know which day you are looking at.

A week that crosses into a new month used to read "31 – Sep 6", with no month on the 31st. The timesheet and the Approvals page now name both months, as in "Aug 31 – Sep 6", and a week that crosses into a new year names both years.

New

Pick agent time repositories from GitHub

Under Settings → Agent time → Repositories, Connect GitHub lists the repositories you share with the Time Trakkr GitHub App. Press Add beside a repository and choose its project and task. Repositories already mapped show where their time goes.

A project's page now has an Agent time repositories section, for admins, showing the repositories mapped to that project and a way to add one.

What this does not do:

  • It does not switch anything else on. Connecting GitHub only lets this page list your repositories.
  • It does not track anything by itself. A repository is tracked only once it is added with a project.
  • Nothing is lost if GitHub is disconnected. A repository already mapped keeps working from its address, and the page notes when GitHub no longer shares it.

Repositories on GitLab, Bitbucket and other hosts are still added by typing their address.

Improved

A returned timesheet now says what needs changing

When you reject somebody's week on Approvals, you are now asked What needs changing? The note goes into their notification, their email and Slack message, and onto their timesheet for that week until they resubmit.

Reopen is now described as what it is. The person is told their week was reopened, not that it "needs changes".

If timesheet approval is switched off, a week submitted before the switch can still be approved or reopened, but not rejected — there would be no way to resubmit it. Reopening it then notifies nobody, because there is nothing for them to do.

Notifications already sent are not changed.

New

Sign the agent time plugin in from your browser

Run timetrakkr login and your browser opens on a page that names the computer asking and says exactly what it will be able to do: report how long your coding agents worked, and read the repository rules that apply to you. Choose a workspace, press Allow access, and the plugin is signed in.

The sign-in appears under Developer → Connected apps as "Time Trakkr CLI on" followed by the computer's name. Revoking it there stops it immediately, and timetrakkr logout does the same from the computer.

What it does not do:

  • It cannot touch your own time. A browser sign-in has the same narrow access as an "Agent time only" token: no reading or changing time entries, timers, reports, invoices or profiles, and no assistant.
  • It only works where Agent time is enabled. For any other workspace the page says so and offers nothing to allow.
  • It only signs in a program on your own computer. A website cannot be given this access.
  • It does not replace the token. A machine with no browser still uses an "Agent time only" personal access token in TIMETRAKKR_TOKEN.
New

Agent time: record it, review it, accept it at an agent rate

Agent time, free on every plan, records how long a coding agent worked on a project, for the person who launched it. An admin switches it on under Settings → Add-ons; until then nothing is recorded.

Agent time is the sum of the agent's turns, from a prompt being submitted to the agent stopping. Time between turns, and time spent waiting on a permission prompt, is not counted.

Review it on your timesheet. Recorded sessions appear in a pending list on your own timesheet's day view, with the project and task, the branch, a one-line summary and the time. Accept a session, change its project or task and then accept, discard it, or accept the whole day. Accepting makes one time entry for the exact time, marked as agent time. It works in a week you have already submitted.

Agent time has its own rates. An admin sets what agent time bills at and costs under Settings → Agent time: a rate for the workspace, with overrides for a client, a project or one agent host, and the option to track agent time without billing it. The rate is fixed on the entry at the moment it is accepted, so a later change never reprices it. With no rate that applies, agent time cannot be accepted, and it never takes the person's own rate.

Accepted agent time is not counted as the person's time. It is listed on their timesheet marked as agent time, beside their day and week totals and never in them, and it is left out of utilization, capacity, reminders and the weekly submit and approve flow. It does count toward the work: project budgets, profitability, uninvoiced time and invoices include it.

On an invoice, agent time has its own lines. Whichever way the invoice is grouped, agent time is on lines labeled "(agent time)", apart from the person's. Each of those lines is rounded once, on its total, to the workspace's rounding step: ten four-minute sessions are forty minutes before rounding, not ten rounded sessions. Agent time that is still pending, or marked not billable, is never on an invoice.

One report, in totals. The Agent work report shows, per person, accepted agent hours by project, the most sessions run at the same moment, the agent's own estimate of the work by hand beside the time it took, and an estimated token cost. An admin sees everybody, a manager their reports, anybody else themselves. What a session was stays with the person who ran it. The report is opened from the pending list or the Approvals page; it is not in the report tabs.

Token cost is an estimate. It prices the tokens a session reported at list price on the day you look, using prices an admin can change under Settings → Agent time. It is not what anybody was charged and is never on an invoice. A model with no price shows "no price set" and a session that sent no token counts shows "not captured"; neither is shown as zero.

The assistant can list and accept your own pending agent time, give the report, and, for an admin, list and set agent rates and model prices. Setting and accepting need a connection that acts as a person: a read-only organization key can read but not change anything.

An admin can turn on auto-accept. It is off by default. When on, agent time from a repository mapped by a workspace rule is accepted for the person once the day has ended in the workspace's time zone, and the entry is marked as accepted automatically. It never accepts a session mapped by a file in the repository, one with no agent rate, one on a project the person may not log time to, or one they discarded. Discarded agent time is deleted after 90 days.

Agent time is measured, so it cannot be raised. An accepted entry can be reduced, moved (which re-prices it) or deleted, and cannot be made longer than the agent measured.

What this does not do yet:

  • Managers do not see what is pending. On the Approvals page they see how many sessions each person has waiting and how much time, and nothing about what the sessions were.
  • Claude Code and the Codex CLI only. Agents that run in the cloud are not supported.
  • No packaged installer. The plugin is loaded from a local copy of its folder.
  • No summary of a session is recorded yet, so a session is named by its agent and branch.

Only metadata is recorded: the repository, the branch, the model, token counts and the timings. Prompts, file contents and diffs are never recorded.

Since then:

  • Auto-accept is on by default since 5 October, unless an admin turns it off, and setup is one page under Settings → Agent time → Setup.
  • Cursor is supported as well as Claude Code and the Codex CLI, since 6 November.
  • The plugins are published and install with one command, since 5 November, so nobody loads one from a local copy.
  • Agents that run in the cloud were not supported when this was written. Later entries say what has changed.
New

Map repositories to projects for agent time

Under Settings → Agent time → Repositories, an admin or owner maps a git repository to a project and task. Paste the remote in any form: the SSH and HTTPS addresses of a repository are the same rule.

A rule can cover every branch, or only branches matching a pattern such as release/*. A branch rule beats the every-branch rule, and the narrower of two patterns wins. The page has a check that shows which rule wins for a remote and branch.

A .timetrakkr.json file in a repository still works, and overrides the rule for that repository.

What this does not do:

  • A repository with no rule is still not tracked. The plugin matches the rules on your own machine and sends nothing, not even the repository's name, for one that is not mapped.
  • You are only told about projects you can log time to. If your workspace restricts members to assigned projects, a rule for a project you are not on does not reach you.
  • It does not move agent time already recorded. A changed rule applies to sessions that start afterwards.

The assistant can list the rules, check what a remote maps to, and set or remove a rule, for admins and owners.

Improved

Turning timesheet approval off now removes the rest of it too

Turning off Timesheets are submitted for approval hid the submit button and left everything around it in place: an Approvals page nothing would ever arrive on, reminders to submit a timesheet that could not be submitted, and a warning on every invoice that its time had not been approved.

With the setting off, those go too:

  • Approvals is no longer in the navigation.
  • The "your timesheet is overdue" and "timesheets are waiting for your approval" reminders are not sent.
  • Invoices do not warn about unapproved time.

What stays. Missing-time reminders still send — they are about logging time, not about submitting it, and they have their own switch. And the Approvals page still opens at its address, with a note saying approval is off, so anything submitted before the change can still be approved or rejected.

Nothing changes for a workspace that has approval on.

Fix

Open an archived person's record in People

Archiving somebody on Team removed them from the People roster, which left no way to open their record — and a leaver's last documents are usually added after they have gone.

Archived people now appear under Archived at the bottom of the roster. The section starts closed, and searching by name finds them without opening it. You can update their profile, record compensation and add documents as before.

Documents can now be added for somebody who has no profile yet. That panel used to appear only once a profile existed.

Archived people are never sent a request for missing details.

Improved

Booking a month of work takes fewer clicks, and the form stops forgetting your dates

Adding bookings on Schedule was built for one booking at a time, and planning a month is never one booking.

  • It opens on the current month. It used to open on tomorrow through five days from today — so on a Thursday it offered Friday to Tuesday, three working days either side of a weekend. That was an offset in calendar days, not a decision anybody made.
  • One click for the usual periods. This week, Next week, This month and Next month set both dates. The dates stay editable, so a booking that starts mid-month is a preset plus one changed date. The same buttons set the dates shown at the top of the page.
  • The form remembers. After you add a booking, the person, dates and hours stay put and only the project clears, ready for that person's next project. The dates also survive a reload.
  • The person picker asks for a team member. It used to read Placeholder (nobody yet), which described the exception as if it were the default. Booking a role before you know who will do the work is still there, as No one yet — book a role.

Two things it will now refuse. The same person on the same project for exactly the same dates cannot be booked twice — a form that keeps its values makes that an easy second click, so edit the existing booking instead. And an end date before the start is caught before it is sent.

If you add a booking for dates other than the ones on screen, the form says so, rather than leaving a booking that appears nowhere.

Improved

Bookings are grouped under each person

The bookings list on Schedule showed one card per booking, in start-date order. Somebody booked on three projects appeared three times, and other people's bookings could land in between — so "what is Jordan on this month" had no single place to look.

Each person is now one group, with their bookings as lines inside it in date order. Beside the name is how many bookings they have and, for the dates on screen, how many hours are booked and how many are still unassigned — or by how much they are over.

Click a name to fold that person away. Placeholder bookings group under the role they reserve, after everybody with a name.

Editing and removing a booking work as before, on the booking's own line.

New

Copy last month's bookings instead of re-entering them

Most months look like the month before. Until now, planning the next one meant entering every booking again.

Copy from another period… on Schedule copies one period's bookings into another. Pick where to copy from — last week, last month, or any dates — and where to copy into, and optionally a single person.

You see the list first. Nothing is created until you have looked at what it will make, booking by booking with the new dates, and you can untick any you do not want.

The dates move the way you would move them. A booking for the whole of October becomes the whole of November, not 1 November to 1 December. A week's bookings stay on the same weekdays. Hours per day, tentative status and placeholder roles carry over.

What it will not do:

  • It never doubles a booking. One that already exists in the target for the same person, project and dates is skipped, so running the same copy twice is harmless.
  • It never changes or removes an existing booking. It only adds.
  • It does not adjust for time off or holidays in the new period. It copies the plan; the capacity list and the time-remaining panel then show you who ended up overbooked.

The assistant can do the same when you are signed in — "copy October's schedule into November" — and it also shows you the list before creating anything.

Fix

Dates everywhere now read your clock, not the server's

A number of places in Time Trakkr worked out "today" from UTC rather than from your own clock. If you are anywhere west of UTC — all of the Americas — those disagree for the last hours of your day, and the app would quietly act on tomorrow's date.

The one worth checking. Marking an invoice sent or paid in full recorded the date for you, and in the evening that date could be the next day. At a month end that puts the payment in the following month, which is the kind of thing that shows up later as a figure that will not reconcile. The same applied to recording a retainer draw, a retainer top-up, and the next issue date on a recurring invoice.

If you have marked invoices paid late in the evening — particularly on the last day of a month — the dates are worth a look. Nothing was lost or double-counted; the date written was simply a day later than the one you were looking at.

Where else it showed.

  • This week on the team page named next week, every Sunday evening.
  • The planning window on Schedule opened a day later than it said.
  • Coverage could show the wrong week entirely.
  • A report's period could be labelled as the current one when it was not, and a report shared with a client could be built from the wrong day's period.
  • Default dates on compensation records, leave adjustments, holiday entries and time-off requests were a day ahead in the evening.

All of these now read the same date your computer shows. Dates you typed yourself were never affected, and nothing already recorded has been altered — this changes what gets filled in from here on.

Server-side timestamps are deliberately unchanged: when something is published or scheduled, that is one moment for everybody rather than a different day depending on who is looking.

Fix

Invoicing and Expenses no longer open with an impossible date range

If you opened Invoice tracked time or Expenses late in the evening on the last day of a month, anywhere west of UTC, both opened with a date range that could not match anything: the start date was already the first of the next month while the end date was still today.

The result was a confident, wrong answer. Invoicing reported no uninvoiced time for a client who had a month of it, and Expenses showed an empty list. Nothing was lost and nothing was double-billed — the figures were never wrong, only the window they were read through — but there was no way to tell from the screen that the range was the problem, and widening it by hand was the only way out.

Both ends of the range now come from the same clock: your own. The dates you set yourself were never affected, and neither was anything already invoiced.

The two date boxes on the invoicing dialog also have names now. They were unlabelled, so a screen reader announced both of them as "date" with nothing to say which was the start and which was the end.

Improved

Notifications are kept for 90 days

Notifications in the bell were kept forever. They are now kept for 90 days and then removed, whether or not you read them. The clear-out happens when you next receive a new notification.

Nothing is lost that was not also sent another way: every notification is emailed too, unless you have turned that email off, and the email is the lasting record.

One thing to know. A notification also stops the same event being sent twice. So something that is still true after 90 days — a project sitting just under its budget — is notified once more when its original notification is removed.

Improved

Report on people's details and systems access, in the assistant or the API

The assistant can now report on employee profiles: job titles, start dates, contact details, linked accounts and which systems each person is recorded as having access to. HR admins can ask about anybody, including people who have left. Everyone else gets their own record only.

It works in the in-app assistant, in Slack, and from Claude through the connector when you connected by signing in.

The REST API has the same report at /api/v1/people, for personal access tokens created with Include People profile data ticked. Existing tokens do not gain it. An organization API key cannot read profiles by either route, and compensation and documents are never included.

Setting up the list of systems has moved to Settings → Systems access, with a button to it at the top of the list in People.

Fix

Personal access tokens carry their owner's role in AI clients

A personal access token is meant to act as the person who created it. Through the REST API it always did. Used as the credential for an AI client over MCP, it was treated as an organization key instead, which reads across the whole workspace. So a member's own token could be used to read things a member cannot see in the app, such as profitability.

Now a personal token over MCP acts as its owner, with the role they hold at the moment of each request:

  • a member's token sees what that member sees;
  • a token whose owner has been deactivated or removed is refused;
  • a token created without the read scope is refused.

What has not changed. A personal token is still read-only over MCP, whatever scopes it has; to log time or change anything from an AI client, connect with OAuth. Organization API keys behave exactly as before. Nothing changes for the REST API.

If an AI client connected with a member's personal token starts answering "only available to managers and admins", that is this fix working.

Improved

Planned vs tracked is grouped by client, and every figure opens its detail

Planned vs tracked on Schedule was a flat list of project names. That stops being readable the moment two clients have a project called the same thing — "Site Build" appeared twice with nothing to say whose each was.

Projects are now listed under their client, in name order, with a subtotal per client. Click a client's name to fold its projects away.

And every figure leads somewhere:

  • A project's name opens the project.
  • Tracked hours open the detailed time report for that project and those dates. A client's tracked subtotal opens it for all of that client's projects.
  • Planned hours open the bookings that add up to the figure, right there in the table.

The numbers themselves have not changed. Planned still counts confirmed bookings only, and work tracked against nothing planned is still shown as unplanned rather than as a percentage.

Clickable figures now carry a dotted underline, here and wherever else a number opens its detail — several already did, and looked exactly like the ones that did not.

New

See how complete each profile is, and ask several people at once

Every profile now has a dial with one segment per field, so you can see what is missing as well as how much. HR admins get it against every name in the roster.

Opening somebody shows what is missing in two lists: what they can fill in themselves, and what only HR can set (job title and start date). Nobody is asked for the second list.

Requests are now remembered. Each profile shows when the person was last asked, by whom and how many times, and the button changes to Resend request once they have been asked. Before, "Request sent" lasted only until the page was reloaded.

To ask several people together, tick them in the roster and choose Request details from N people. You are told how many emails went, who needed nothing, and who could not be emailed.

The email itself is short: a friendly ask and a Complete my profile button. It does not list what is missing; the profile shows that when they arrive.

It does not send reminders by itself. A request goes out only when an HR admin sends one.

Improved

Every figure on a project page now opens its detail, and you can see which ones

Some figures on a project page already opened the entries behind them. You could not tell which: a number that was a link looked exactly like one that was not, and the arrow that opened a task row was small enough to miss.

Clickable figures have a dotted underline. And now all of them are clickable:

  • Total hours, and Billable / Non-billable separately, open the detailed time report.
  • Budget remaining opens the entries counted against the budget.
  • Internal costs opens the time report for Time and the expense report, narrowed to this project, for Expenses.
  • Invoiced opens the Invoices tab; Uninvoiced and New invoice open the uninvoiced list rather than the invoice overview.
  • Each table's Total opens exactly the entries it adds up.
  • Clicking a bar on the chart shows that week or month below.

Rows open by clicking the name. A task opens onto who logged it, as before, but the whole name is the control and the arrow is large enough to read as one. People on the Team tab now open the same way, onto the tasks their hours went to.

One fix that came with it. A figure shown under All time used to open the report on the current month — so three hours logged in May opened onto an empty October. The link now carries the whole range.

Nothing about who can see what has changed. A member following one of these links still sees only their own entries, and cost figures stay managers-only.

Fix

A report's custom timeframe no longer opens on an error

On the detailed time and expense reports, switching the timeframe to Custom filled the end date with the last day of the current period — the 30th, say — and immediately flagged it, because the box only accepts today or earlier. The form opened invalid, over a date you had not typed.

The end date now opens on today when the period has not finished. A period that has already ended opens on its own last day, as before.

Reports and comparisons over This month, This week and the other presets are unchanged: they still run to the real end of the period, which is what lets "this month against last month" compare the same number of days.

Improved

Schedule lists start closed, with one click to open or close them all

Planned vs tracked and the bookings list on Schedule opened with every client and every person fully expanded. In a workspace with thirty clients that was a long page to scroll before you had asked to see anything.

Both now open closed: clients with their planned and tracked totals, and people with how many bookings they have and their booked and unassigned hours. That is the overview; opening a row is asking about one of them.

  • Click a client or a person to open just that one.
  • Expand all at the top right opens every one; once they are all open it reads Collapse all.
  • When you add a booking, that person's group opens by itself, so you can see the booking land.
New

Track which systems each person has access to

An HR admin can switch on Systems access from Settings → Systems access, and list the systems worth tracking: Google Workspace, GitHub, a password manager.

Every profile then has a tick box per system. Tick it when access has been given, clear it when it has been taken away. Each box shows when it last changed and who recorded it.

When somebody is archived with boxes still ticked, their profile says so, and the setup page lists everyone in that position with the systems outstanding.

It is a record you keep by hand. Ticking a box does not grant access and clearing one does not remove it — nothing is connected to the systems themselves. Only HR admins can see or change it.

It is off until an HR admin switches it on.

Improved

The team-member picker on Schedule offers search from six people

The pickers in the app add a search box once a list reaches eight entries. For the team-member picker on Schedule that left a six- or seven-person workspace scanning the list by eye every time it booked somebody.

That picker now offers search from six people. Every other picker is unchanged.

Improved

Every time entry now says how it was recorded

Each time entry now records where it came from: typed into the timesheet, a timer, the assistant, a connected AI client, Slack, the API, an import or a calendar draft. The Detailed report shows it in a new Source column, and the CSV export carries the same column.

Time you log by talking to the assistant, through a connected AI client or from Slack is still your own time. The source only says how you wrote it down.

Entries from before this change are marked from what was already known about them: an entry a timer created is Timer, an imported entry is Import, and one confirmed from a calendar draft is Calendar. Everything else is Typed, including entries that were in fact logged through the assistant, Slack or the API, because nothing recorded that at the time.

For API users: POST /api/v1/time_entries now refuses a field it does not know with a 400 that names it, where it used to ignore the field and create the entry anyway. If an integration starts failing, the response says which field to remove.

New

See who is away this month, and decide from the same screen

Time off has had a list of requests since it shipped. A list answers "who asked for what". It does not answer the question you usually have, which is "does anything break that week".

People → Time off → Time off calendar lays the month out: one row per person who is away, one column per day. You can see at a glance that two of the three people who know the deployment are off the same week, which is not something a list of requests will ever tell you.

The approve and decline buttons are on that screen, under the grid, and that is deliberate. Deciding one request without seeing the week around it is how a team ends up with nobody covering.

Reading it

  • Solid is approved. Striped is requested and not yet agreed.
  • Half filled is a half day. A shaded column is a weekend.

The solid-versus-striped distinction is the one worth remembering: pending time off is not time off yet, and planning cover around it is the mistake the stripes exist to prevent.

Who appears

The same rule as everywhere else in People, enforced by the database rather than by the page: a member sees their own time off, a manager sees themselves and the people who report to them, an administrator or owner sees everybody.

Only people who actually have time off in the month get a row. Somebody with nothing booked is not listed — a wall of empty rows buries the answer. When nobody you can see is away, the page says so and names the month.

Colors

Each leave type now has a color, set under People → Time off → Configure next to the type's name. Until you choose one, each type is given a consistent color of its own, so the calendar reads properly before anybody configures anything.

What it is not

It does not show public holidays or working patterns — it shows time off that was requested. For "who was expected to work and logged nothing", Reports → Coverage already does that, and it does account for holidays and non-working days.

New

See how much of somebody's time is left while you book it

Planning somebody's month meant adding a booking, scrolling up to the capacity list to see what was left, and scrolling back down. The answer is now under the form, for the dates you are booking rather than the dates on screen.

Pick a team member and you see:

  • What they have unassigned — their capacity for those dates, less approved leave, less what is already booked.
  • What this booking would leave. Change the hours a day or the dates and it updates. If the booking would take them past their capacity, it says by how much.
  • How much of the project's budget is not yet booked, when the project has an hours budget.

Three things it deliberately does not do. It does not stop you overbooking somebody — sometimes that is the plan, and the point is to do it knowingly. It does not count tentative bookings, which are pipeline rather than commitment. And the project figure compares the budget with bookings, not with tracked time: it tells you how much of the budget has been planned, not how much has been spent, which is what the project page is for.

The capacity list at the top of the page now shows the unassigned hours for each person too. You can also ask the assistant — "who has room in October?" — when you are signed in; it is not available to a read-only API-key connection.

New

Timesheet approval can be turned off

Not every team approves timesheets. Until now the Submit week for approval button sat on every timesheet whether or not anybody was ever going to press it.

Under Settings → Preferences, untick Timesheets are submitted for approval. From then on:

  • the submit button is not shown on anybody's timesheet;
  • a week cannot be submitted;
  • timesheets are not auto-submitted at the deadline, even if that was switched on before.

Nothing else about tracking changes. Time still appears in every report and can still be invoiced — approval was never required for either.

What it does not do. It leaves alone anything already submitted or approved: those entries keep their status, and approved ones stay locked. The Approvals page is still there, and so are the reminder emails about missing time. This removes the submit step, not the history.

It is on by default, so nothing changes for a workspace that does not touch it.

September 2026

New

A nudge that names the fields, and can reach 100%

People → My profile used to say "your profile is 80% complete, 2 fields left" without saying which two. It now lists them, and clicking one takes you to that field.

It can reach 100% now

Two of the tracked fields — job title and start date — are set by an administrator rather than by you, on purpose: start date drives leave accrual, so it is not something anybody should be able to set about themselves. They were still counted in your score, which meant a profile you had completely filled in still read 80% and the nudge asked for something the form could not give it.

Your score now counts only the fields that are yours. Administrators still see the full picture.

Asking somebody for theirs

Administrators get Request missing details on a person's profile. It emails them naming the fields still outstanding, with a link straight to the page.

It only ever asks for fields that person can actually set, and it refuses to send if there is nothing missing — a request asking for nothing reads as a mistake to whoever receives it.

It is still only a nudge

An incomplete profile has never blocked tracking time and still does not. The missing fields sit beside the ordinary form rather than in a wizard you have to step through, because a wizard is the thing that tends to become a gate.

New

The bell now covers approvals, backups and overdue invoices

When the bell shipped, only budget alerts actually produced anything. The other switches were real and nothing wrote to them. They all work now:

  • Someone submits a timesheet you approve.
  • Your timesheet is approved, or returned for changes.
  • A scheduled backup starts failing.
  • An invoice you raised goes past its due date.

Each one reuses the rule that already decides who gets the corresponding email, rather than a second rule of its own — so the two channels cannot disagree about who should hear.

A workspace-wide switch

Settings → Preferences → In-app notifications, for administrators and owners. Each person still chooses which kinds they want on their own profile; this decides whether the workspace uses the feature at all, which an administrator introducing it had no way to say before.

Switching it off stops notifications being recorded, not merely shown. A workspace that turns it off does not accumulate a backlog that arrives in full the day somebody turns it back on.

The approval email that was promised and never sent

Email me when a timesheet is approved has been a preference for a long time, and nothing sent it. It does now. It stays off by default — an approval needs nothing from you, unlike a returned week — so this only reaches people who had already asked for it and were getting nothing.

An overdue invoice is in-app only, and that is a decision rather than a gap: nobody asked for invoice nags by email, and adding them would put a recurring message in every administrator's inbox they never requested. The bell is the right weight for it.

Everything else on the list emails exactly as it always did.

Backups tell you even without email set up

The failing-backup notice used to return early when a workspace had no mail transport configured. The bell does not depend on that any more, which matters most for exactly the workspace that has no email set up and would otherwise never learn its backups were broken.

And in your browser, if you want

Open the bell and tick Also show these in this browser. New notifications then pop up on your desktop while Time Trakkr is open in a tab you are not looking at — the case where something arrived and you missed it because you were in another window.

It is per device, because the browser permission is. Nothing pops up for what you already had when you opened the tab, and nothing pops up while you are looking at the page.

It does not reach you when Time Trakkr is closed. That needs a different mechanism, and it competes with the emails rather than adding to them, so it is a separate decision rather than a missing piece of this one.

Fix

Planning work now has the same project and people pickers as everywhere else

When you add a booking on Schedule, the Project and Person pickers were still plain dropdowns. Every other project picker in the app had already moved to the searchable one that groups projects under their client; this screen was the last one that had not.

That mattered most where it was least visible. Several clients can have a project with the same name — "Development Monthly Retainer" is a real example — and a flat list showed them as identical rows with nothing to tell them apart. The client was always in the data; only this screen was discarding it.

Both pickers now behave like the rest of the app:

  • Projects group under their client, with the client shown on the control once you have chosen one, so the booking you are about to create is unambiguous.
  • Both can be searched once the list is long enough to need it. Below that the list is short enough to read, so the search box stays out of the way.
  • Placeholder bookings are unchanged. Leaving the person as Placeholder (nobody yet) still reveals the role picker, so you can reserve capacity for somebody you have not hired yet.

Nothing about existing bookings changed — this is the picker only.

New

Tell us when the assistant gets it wrong

Every assistant answer now has a thumbs-up and a thumbs-down under it. Thumbs down also offers a box for what was wrong, which is optional and is by far the most useful thing you can send us.

Clicking the same button again withdraws your rating rather than reversing it, so a mis-click leaves nothing behind. Your rating is yours — nobody else in your workspace sees it, administrators included. Deleting a conversation deletes its ratings with it.

What we review, said plainly

This is on unless your administrator turns it off, so it deserves a straight account rather than a line in a policy.

We review the question you asked, which tools the assistant chose, and your rating. Tool names are part of Time Trakkr's own code — whether it reached for the time report or the approvals list — so they say nothing about your business.

We never review the answer, and never anything a tool returned. An answer states your figures in sentences, which makes it the most revealing part of the exchange; what is useful to us is that an answer was rated unhelpful, not what it said. Your hours, clients, projects, people and amounts are removed before anything leaves your workspace, and the code that removes them is tested against exactly that. Your workspace is identified by an internal id, never its name.

Turning it off

Settings → Assistant → Helping improve the assistant. Administrators and owners.

Switching it off applies immediately to everything already collected, not only to what happens next. There is no waiting period, because the setting is checked at the moment anything would be read rather than stamped on each rating as it is given.

You can also ask the assistant whether questions here are reviewed, and it will tell you which way the setting is set. It deliberately cannot change it — a privacy decision should not be made by asking the thing it governs.

One thing this setting does not do

It does not stop your conversations being saved. They are saved either way, because the assistant cannot follow a thread from one message to the next without them — that is what lets you ask "and the month before?" and be understood.

What the setting decides is whether those questions and ratings may be reviewed by us. We would rather say that than offer a switch whose label implies something the product cannot do. Conversations remain yours to delete at any time from the assistant's sidebar.

New

A project now tells you it is near its budget

Setting Send email alerts if project exceeds 80% of budget did exactly one thing: send an email. If you had not set it up, or it was filtered into a folder, you could open a project every day and never learn it was over. The figures were on the page; noticing meant doing the arithmetic yourself, which nobody does on a project they are not already worried about.

The same percentage now marks the project where you are working.

  • On the project page, a band above the figures once it is at or past the threshold: how much of the budget is used, how many hours are left or over, and whether that is against this month or all time.
  • On the Projects list, a Near budget or Over budget tag beside the name.
  • Above the list, a count of the projects needing attention. Select it to see only those; select it again to go back.

What it will not do

Nothing appears when nothing is wrong. There is no permanent "0 projects over budget" line. A message that is always there is one people learn to skip, including on the day it says something else.

Over budget still shows when email alerts are off. Switching the alerts off means you do not want the email — not that the project should stop telling you it is over. With them off there is no early Near budget warning, because the warning line and the over-budget line become the same number.

Who can see it has not changed. A member sees a project's budget state only where Show budget progress to everyone on this project is ticked, hours only, never money. The assistant follows the same rule.

You can also ask the assistant "which projects are near their budget?" without naming one first.

Improved

Custom on a report now opens where you already were

Reading September on the report builder and choosing Custom set both date boxes to today, so the report you were reading was replaced by a single day and you had to type both dates before it said anything useful again.

Nobody picks Custom to mean "today". They pick it to adjust the span already on screen, so that is where it now starts: choose Custom while reading September and you get 1–30 September, ready to widen or narrow.

Switching to another period and back re-seeds from that period. A range you typed earlier is not restored, because it would put figures on screen matching neither the period you were just reading nor anything visible — which is more surprising than starting again, not less.

The project page has worked this way for a while. This is the same behavior in the other place that asks.

New

A bell for the things Time Trakkr needs to tell you

There is now a bell beside your name in the sidebar. It shows the things Time Trakkr needs to tell you while you are actually in it: a project reaching its budget, a timesheet waiting on you, a backup that has started failing.

It does not replace the emails

Everything on the bell is also emailed, exactly as before. They are two channels for one event rather than alternatives, and each has its own switch. Turning off an email does not take the notification off the bell; turning off the bell does not stop the email.

That is the whole design decision, and it is worth being plain about because the opposite is the common one. Email reaches you when you are not in Time Trakkr. The bell reaches you when you are. Replacing one with the other would mean choosing which of those you would rather lose.

The badge does not clear itself

The number counts what you have not marked read, and opening the panel does not empty it. A badge that clears because a tab was left open has told you something is handled when nobody has looked at it.

Clicking a notification marks that one read — following it to the project or timesheet it is about is reading it. So does the tick beside it, and Mark all read at the foot of the panel. The count stops at 9+.

Choosing what you get

Team → your name → Notifications, under In the app. Six switches, one per kind of notice, all on to begin with.

They deliberately do not mirror the email preferences above them. Those were written for different things — "email me if timesheets are submitted for people I manage" and "…for projects I manage" are two emails about the one event that appears as a single line on the bell.

What it is not, yet

It does not reach your browser or your phone. Notifications appear only while you have Time Trakkr open in a tab. Browser notifications are a separate piece of work and are not built.

Nobody else can read yours. A notification belongs to a person rather than to a workspace. An administrator cannot see them, there is no team-wide view, and that is enforced in the database rather than by the page choosing not to show them.

Nothing on it involves money. A budget notification names the project and the percentage it has reached, never an amount — the same rule the weekly project rollup email already follows.

Budget alerts are wired up now. Approvals, backups and overdue invoices follow as each of their emails is connected to the same rule, so that the two channels can never disagree about who should hear.

You can also ask the assistant — "what have I missed this week?" It reads your own notifications and nobody else's, and it cannot mark them read.

Improved

A teammate's week now reads day by day

Opening View on somebody's week in Approvals shows every entry they logged. The days were already separated — each one headed with its date and its own total — but the heading was set in small, faint, grey type while the entries below it were full-size and bold.

So the divider was the quietest thing in the table, and eighteen entries read as one undivided list. The information was there; it just carried no weight.

Each day is now a filled band: its date and its total in full contrast, with a rule above and below. Nothing has moved and nothing has been added — the same headings, the same totals, in a weight that matches the job they are doing.

Fix

Big breakdowns from the assistant no longer cut the totals short

Ask the assistant to break hours down by project, person or day and it returns a row per group. That was capped at 100, which an ordinary agency passes without trying — one report came back "truncated at 100 of 105 rows".

Two things were wrong with what happened next.

The cap was too low. It is now 500. A hundred-odd clients or projects is normal, and a limit reached by asking an everyday question is a limit in the wrong place.

The totals were only partly protected. Total hours were always summed across every group, capped or not. Billable hours, revenue and cost were not — so when the cap bit, the assistant had nothing to quote but the rows in front of it, and said so: "the grand total is based on the 100 returned." It was being honest about a number it had not been given.

Every total now spans everything that matched. When a result is still capped the assistant says so, and shows the largest groups rather than the first alphabetically — but the total beside them is the real one.

The same fix, everywhere else it was needed

Expenses, submitted timesheets and estimates were capped the same way and had no totals at all.

  • Expenses and estimates now report a total per currency. Deliberately not one number: both carry their own currency, and adding dollars to euros produces a figure that looks like an answer and is not one.
  • Timesheets awaiting approval now report total hours, total entries and how many people are waiting — which is what the question usually is.
Fix

A corrected time entry no longer stops the week loading

If somebody corrected an over-log in Harvest by entering negative hours, that entry came across on import and the whole week refused to open. Not the one entry — the entire timesheet for that week, for that person.

Thirty-three such entries were blocking eight people across three organizations, some of them for months.

The week now loads, the correction is shown with a minus sign, and it counts towards the day and week totals the way Harvest counts it. So the figures match what Harvest reports, rather than being too high by the amount somebody corrected.

A fractional duration is still refused. That is arithmetic that went wrong somewhere rather than a number anybody typed, and quietly rounding it would hide the fault that produced it.

New

Filter the projects list by who manages them

The projects screen filters by status and by client. It now filters by manager too — choose a person and the list narrows to the projects they manage.

It reads the same assignment that already decides who may see a project's report, so there is nothing new to set up: anybody marked as managing a project on the project's team appears in the dropdown.

A project with two managers — a handover, usually — shows under both. And if nobody in your organization is marked as managing a project, the filter is not offered at all rather than sitting there empty.

New

See what a re-import kept your version of

A re-import already protects your edits. If you renamed a client here and the name also changed in Harvest, the import keeps your name rather than overwriting it.

What it did not do was tell you. The detail went into a server log and you got a number, so "did the import overwrite my change" had no answer you could see.

Integrations → Kept your version now lists them: which client, project or task, which field, your value and the source's, and when it was noticed.

Nothing here needs doing

Every row is a decision already taken, and taken the right way round. The list exists so you can disagree, not because something is waiting on you.

  • Keep mine records that you looked and agreed.
  • Use the source value writes the source's value over yours and stops that field conflicting in future — it starts tracking the source again.

Re-importing weekly does not pile up copies: a field that keeps disagreeing stays one entry, with the date it was last seen. A fresh disagreement after you have settled one is a new entry, because it is a new event.

Why it is worth glancing at

One row is somebody's edit being protected. The same row appearing every week means an import is quietly declining the same change indefinitely, which usually means the two systems disagree about something that should be settled in one of them.

New

Managers can correct a teammate's timesheet

Opening a teammate's timesheet used to show it read-only. You could see that somebody had logged 28 hours on a Tuesday because they forgot to stop a timer, and you could do nothing about it except ask them to fix it themselves.

Managers, administrators and owners can now edit those entries directly: change the hours, move time to the right project, correct the notes, add an entry somebody missed.

Every change is recorded

Editing your own timesheet is the ordinary business of the day and is not logged. Editing somebody else's is, every time, with who made the change, whose time it was, and what moved — "8:00 became 7:30" rather than just "something changed".

That is the part worth knowing about. The screen being read-only was never really the protection it looked like — the underlying API has always allowed a manager to do this, so it could already happen without leaving a trace. Now it leaves one.

What has not changed

Approved and invoiced entries stay locked for everybody, and moving an entry into or out of a submitted week is still refused.

Timers are still yours alone. You can record how long something took on a teammate's behalf, but you cannot set a clock running in their name — they would be left with a timer they never started. Submitting a week for approval is still theirs too; that is an assertion about their own work.

Who can do this follows the same rule as approvals: a manager sees the people they approve for, and administrators and owners see everybody.

Jumping to a date

The date on the timesheet is now the control that changes it. Select it to pick a day from a calendar instead of stepping the ‹ › arrows one day at a time — which was fine for yesterday and useless for last month.

New

Share a report with a client, without giving them a login

Share now covers both halves of the question. With a teammate hands over the report itself, which re-runs under their own access. By email sends somebody outside the organization a page they open in a browser with no account and no login.

They are one control because they answer the same question, and they stay clearly labelled because they disclose very different things.

Build a report, choose a client, and send it. You get a link your client can open with no account and no login. It is a snapshot: the figures are fixed at the moment you publish, so a link you send in September still shows September's numbers when somebody opens it in March.

That is deliberate. A link to a live report would keep re-running, quietly picking up projects started since, people hired since, and work you never meant to include. When you want a client to see fresh figures, you publish again.

What the client sees

Hours, for the period, for that client. Nothing else.

No rates, no amounts, no costs. A link can be forwarded in a way a screen cannot, so money stays out of it entirely — the same rule the weekly project rollup email already follows.

Names are not included unless you tick the box. Left alone, the client sees totals by project or task. Tick Show who did the work and it names individuals and their hours. It is off by default because telling a client who did what is a decision worth making each time, not one to inherit.

Who can open it

Sending by email asks you to choose.

Anyone with the link is the simpler option: a URL that works for whoever holds it. You can still have us email it to people — who is sent the link and whether opening it needs proof are now two separate choices.

Confirm their email address is the default. Each address you type is sent the link. When one of them opens it, they enter their email address and get a one-time link that confirms it — that link works once and expires in half an hour. Forwarding the URL to somebody not on the list passes on nothing; they get the same form, and an address that is not on the list produces the same response as one that is, so the page cannot be used to find out who you work with.

The list is what decides, and it is re-read every time the page is opened. Taking somebody off it takes effect immediately rather than whenever their browser happens to forget them.

It is worth being plain about what this is: confirmation that the reader holds the address you sent it to. Somebody who can read that person's email can open the report. It is not a password.

Seeing what you have sent, and taking it back

Reports → Shared lists everything sent to a client: which client, which period, whether it is gated, who it went to and — for a gated link — who has actually opened it, which is not the same question. Each one has Copy link and Revoke.

An open link deliberately reports nothing about its readers. It cannot know, so it says so rather than showing you a zero.

Revoke stops a link working. Anyone opening it after that gets nothing — no page, no hint that there ever was one. What you published stays recorded, so you can still answer "what did we send them, and when".

Sending to a client is available to managers, administrators and owners, on summary reports. Detailed reports — the line-by-line list of individual entries — cannot be sent this way; that is a different decision from sending totals.

Where an option is unavailable the dialog now says which and why, instead of the button quietly not being there.

Fix

Budget alerts now say which client, and which period the hours cover

A budget alert told you a project had crossed its budget and left you to work out the rest. Two things were missing, and together they could lead you to a confident wrong conclusion.

It did not name the client. Plenty of projects are called "Monthly Retainer", so the email identified the work only if you happened to have one of them. It now shows the client alongside the project.

It did not say what the hours covered, and for a budget that resets each month it counted every hour ever logged against a single month's quota. A 20-hour monthly retainer with a year of history behind it looked hundreds of hours over and read as permanently blown, which is not something the number can be trusted through. A recurring budget is now measured over the current month, the same figure the project page has always shown, and the email names the period it is reporting — "Logged (September 2026)", or "Logged (all time)" for a budget that does not reset.

A note about hours possibly belonging to the previous cycle has gone with it. It existed to hedge a figure that was not period-scoped; now that it is, the caveat would be wrong.

Two smaller corrections came with this:

  • A recurring budget now alerts once per period rather than once ever. If you crossed a monthly budget in one month and heard nothing in the months after, that is why.
  • If your organization counts only billable hours toward budgets, the alert now respects that. It previously counted everything, so the email and the project page could disagree.
Fix

Dropdowns inside a dialog are no longer cut off

Opening a picker inside a dialog — choosing a project on a new time entry, for instance — showed a list cut off at the edge of the dialog, with the rest reachable only by scrolling a small strip.

This started when dialogs were made scrollable so a long one would stop running off the bottom of the screen. Making a dialog scroll also makes it clip anything that tries to escape it, and the dropdown was one of those things. The list now opens above the dialog rather than inside it, so a long list of projects and tasks is visible at full height again.

Both now work: a long dialog still scrolls, and a dropdown inside it still opens to its full size.

Fix

Re-importing no longer overwrites a rate you set yourself

A person's billable rate on a project is what their time is invoiced at. Re-running a Harvest import overwrote it from Harvest's own rate every time, so a rate you had set deliberately could be replaced without any sign that it had happened — and the next invoice was built on the replacement.

A rate you have set here is now left alone. Where no rate has been set, Harvest still fills it in, so an import continues to be useful for the assignments you have not made a decision about.

One consequence worth knowing: once a rate exists here, changing it in Harvest will not change it here. If you are still keeping rates up to date in Harvest, update them in both places, or change them here and treat this as the record.

New

Compare a report against the period before it

Any report built over a period can now be read against the one before it. Tick Compare to previous period on the report builder, on Time, or on Profitability, and every row gains a change column: what it was, what it is now, and the difference.

You do not pick the second period. It is derived from the one you are already looking at — last month against this month, last quarter against this quarter, last week against this week. A custom range compares against the same number of days immediately before it. The caption above the table always names the period it compared against, so a number never appears without saying what it is a change from.

Three things it deliberately does rather than the obvious alternative:

A comparison against a month still running is cut short, not flagged. On the 18th, comparing September against August would be eighteen days against thirty-one, which always shows a fall — and the fall would be believed. So the previous period is trimmed to the same number of days that have elapsed, and the caption says so. It is a like-for-like comparison rather than a warning you have to remember to apply.

A project that has stopped keeps its row, marked gone, showing the full drop. Leaving it out would let every remaining row show growth while the total fell, with nothing on screen explaining why.

Something with no hours last period reads new, not +100%. A percentage against nothing is a number with no meaning.

What it will not do: compare two periods you choose independently — the second one is always the one before the first. It is also unavailable when you group rows by date, because last period has different dates, so every row would match nothing and read as new; the tick box explains this when it is unavailable. Comparison applies to the summary view, not the detailed entry list.

Saving a report saves the comparison with it, so a saved report that compares opens comparing. Exported CSV carries the previous period's own figure alongside the current one rather than the difference, so you can do your own arithmetic.

Fix

Re-importing no longer removes your project managers

"Manages this project" on a project assignment is not just a label — it lets that person see everyone's hours on the project and pull its reports. Running a Harvest import again overwrote it from Harvest's own project-manager flag, so somebody you had given access to could quietly lose it, with nothing to connect the loss to the import.

Harvest has no equivalent of that permission, so it is no longer treated as the authority on it. A brand-new assignment still picks up Harvest's flag when it is first created — there is nothing of yours to preserve on a project nobody has been assigned to yet — but from then on the tick is yours.

Other details still come from Harvest on every import, including project and task names and rates. If you have changed one of those here and want it to stay, tell us — we are working out how conflicts should be presented rather than silently resolved.

Fix

A report with an unfinished date range now says so

Choosing Custom on the report builder and then clearing or retyping one of the dates left the report showing whatever it had last worked out, under dates that no longer matched it. A date box reports every value you pass through while typing, so a half-finished date was being treated as a real one.

The builder now waits until both dates are complete and in order, and says so rather than leaving stale numbers on screen. A range that runs backwards is treated the same way.

Fix

The assistant can read your workspace settings

Asked whether notes on a time entry could be made mandatory, the assistant searched the help articles, found nothing, and answered that the feature did not exist — then suggested workarounds. The setting was on the Settings page the whole time.

It had no way to read configuration at all, so the only evidence it could find was a missing help article, which says something about the documentation rather than about the product.

It can now read your workspace settings directly: time tracking rules, rounding, week start, duration and date formats, approvals and locking, visibility, invoicing defaults, security posture, and which modules and add-ons are switched on. The same is available over MCP and the Slack command, and it reads exactly what the Settings page shows you — no more, no less.

It has also been told that silence in the help center is not evidence a feature is absent, so an undocumented capability gets "I could not find it" rather than a confident denial.

Two related fixes shipped alongside:

Requiring notes now holds over the API too. The rule was enforced in the timesheet, on the timer and in the assistant, but not on /api/v1 — so a workspace could switch it on, see it hold everywhere it looked, and still accept note-less entries from an integration. An edit can no longer clear a note the rule requires either. Entries saved before you switched the rule on are untouched, and an edit to one of those that does not touch the notes still goes through.

Two new help articles, covering workspace settings and notes on time entries.

Fix

Links in answers look like links, and a failed question is no longer lost

Links you can see. When the assistant told you where to change a setting, the place it named really was a link — it just looked exactly like the rest of the sentence. Links inside an answer are now colored, underlined, bold, and carry a small arrow marking that they go somewhere. The same styling reaches the blog and What's new, which had the same problem.

A question that fails is kept. If a turn failed — no AI key configured, an upstream outage, a fault of ours — the reply was an error and your question vanished with it. Nothing was recorded, so nobody could see what had been asked or that it had failed at all.

The conversation now keeps both the question and the explanation you were shown, so it reads back the way you experienced it and you can pick the thread up rather than retyping. Real faults are reported to us as well. Two things deliberately excluded from that: a workspace that has not configured a key, and an upstream rate limit. Both are ordinary states rather than breakages, and alerting on them would bury the failures that matter.

Improved

The assistant links you to settings, and you can rename a chat

No more internal field names. Asked about requiring notes, the assistant answered correctly but showed its working — quoting notesRequiredOnEntries: false as evidence. That is the name of a key in a payload, not something that exists in the app you are looking at. It now says "Time entry notes is set to Optional", using the label on the screen, and keeps code formatting for things you would actually type, like an API field you are integrating against.

Settings are links now. When an answer points you somewhere — "turn it on under Time entry notes" — the place is a link that takes you to that row on the Settings page, showing its current value with Edit beside it. The link is built from the same label the answer quotes, so the two cannot drift apart.

Rename a chat. Titles are generated from the first message, so a conversation that wandered stayed filed under whatever it opened with. Hit the pencil on any chat in the sidebar, or double-click its name, and type a new one. Enter saves, Escape cancels. A chat someone shared with you keeps their name for it — that is what they will be looking for.

Improved

Sharing is easier to find, and project filters follow the client

Sharing a chat. The control was a small diagonal arrow that appeared when you hovered a conversation in the sidebar — and a diagonal arrow usually means "opens in a new tab", so it pointed at the wrong idea as well as being hard to spot. It is now the standard share mark, and there is a labeled Share this chat button above the conversation you are actually having.

Sharing a report. Share lived only on the Saved reports list, so from inside a report — the moment you decide it is worth sending to someone — there was no way to do it. The button now appears on the report itself, for reports you own. A report somebody shared with you is still not yours to pass on.

Project filters follow the client. On the detailed time report, choosing a client left the Projects list showing every project in the workspace, grouped by client, which made the second filter harder to use than no filter at all. It now lists only that client's projects, and drops the client headings that made it look like a client search. Changing the client clears a project that does not belong to it, rather than leaving two filters that contradict each other and a report that returns nothing.

Fix

Reviewing a week shows a total per day, and scrolls

Opening View on Approvals to read somebody's week showed every entry, a total for the week, and no way to reach the bottom of a long one — the dialog ran off the screen and would not scroll. Any long dialog in the app had the same problem; this is just the one you meet most often.

Dialogs now scroll when they outgrow the window.

The same dialog now totals each day beside its heading, and repeats the week's total at the foot of the table where it stays in view after scrolling. Checking that somebody's Tuesday adds up no longer means adding it up yourself.

Improved

Make someone a manager across all of a client's projects

Assigning someone across all of a client's projects could set their role but not whether they manage those projects — so making somebody a project manager for a client meant ticking the box on every project by hand.

The panel now offers Leave as they are, Make them a manager and Remove manager, alongside the role. It leaves each project alone unless you choose otherwise, so changing somebody's role never quietly changes what they can see.

It is also easier to spot. It was a line of grey text among other grey text, and people were missing it entirely.

Managing a project lets somebody see everyone's hours on it and report on it. It still does not give them rates or margin — that has not changed.

New

Expand a task on a project to see who logged it

The Tasks table on a project gave you a total per task and no way to ask the obvious next question. Finding out whose hours made up "Project Management" meant leaving the page and building a report.

Each task row now has a caret. Open it and you get a line per person who logged time against that task: their hours, their cost if you can see costs, and a chip marking anyone who manages the project or whose account has been disabled. The hours are a link, as they are everywhere else on this page — it opens the entries behind that one person's figure on that one task.

The breakdown follows whatever date range the table above it is showing, including a custom range, so the people always add up to the task total beside them rather than to some other period.

It will not show you anybody you could not already see. The expansion is filtered exactly as the table above it is, so a project manager sees the people on the projects they manage and nothing wider.

Improved

Find a person by name on Team and People

Team could be filtered by department, group and status; People could not be filtered at all. Neither could be searched by name, so finding one person meant reading the whole list.

Both now have a search box. It matches on name, email, job title and work role, so whichever part of somebody you remember will find them, and it ignores accents — searching "Jose" finds "José". Typing more than one word narrows rather than widens: "jor lee" finds Jordan Lee.

Fix

A re-import no longer mislabels time that moved project

When you re-run an import, we fill in the task on any entry an earlier run left blank. If you had since moved that entry to a different project in Harvest, we worked out the task from its new project but wrote it onto the entry where it still sat — so the entry ended up labelled with a task belonging to somewhere else. On the project page that showed up as the same task name appearing twice in the list with different hours.

A re-import now leaves those entries alone rather than labelling them wrongly. Nothing is moved between projects: an entry that changed project in Harvest keeps the project it has here, because re-attributing tracked time is not something an import should decide on its own.

If you have entries already affected, this release does not change them — get in touch and we will sort them out with you, since the right answer depends on where the time should actually sit.

Fix

Hours read the same way everywhere

Settings lets you read hours as 7:30 or 7.50, and about half the app honoured it. The rest showed decimal regardless — so a project could report 78.40 total hours with "Billable 78:24" written directly underneath. Every place hours appear on screen now follows your setting, and so do Slack replies and budget alert emails.

The more important half is typing them in. Only one field in the app accepted the HH:MM it was showing you. Everywhere else expected a decimal, and three of them failed silently rather than telling you:

  • Typing 1:30 into a project budget saved it as zero, wiping the budget with no error.
  • Typing it into a task or team-member budget cleared that budget when you clicked away.
  • Typing it into a calendar draft emptied the box, so the time you entered simply vanished.

Every duration field now takes 1:30 or 1.5, whichever you prefer, and shows the value back in your own format. Clearing a budget still works — it just needs an empty box rather than a typo.

Exports and the API are unchanged and stay decimal on purpose: a spreadsheet can add up 7.50 and cannot add up 7:30.

Fix

Only the owner changes the subscription

Owner and Administrator did the same things. They now differ in one place: the subscription. Administrators can still see which plan you are on and its standing — that is useful, and asking someone else just to find out is not. Changing the plan is the owner's.

Nothing else moves between the two roles.

Improved

Report a project over any date range

A project's hours could be read over all time or over a whole calendar month, and nothing else. If your reporting period does not line up with a month — a sprint, a statement of work, the fortnight a client is asking about — the answer was not on the page.

The range selector now has Custom range. Pick it and two date boxes appear; the hours, the task and team tables, and the total at the top all follow the dates you choose. Switching to Custom starts from the period already on screen, so you can widen or narrow what you were looking at rather than starting over.

Export now says what it covers. The file used to be named after the project alone, so two exports of different periods arrived on your disk under the same name and looked identical once opened. The range is now in the filename and in a column inside the file.

One thing worth knowing: for a budget that resets every month, the "Budget remaining" card keeps reporting a calendar month even while the rest of the page shows your custom range. A monthly allowance has no meaning over an arbitrary span of days, and we would rather tell you which month we mean than invent a number. The card says which month it is showing.

Fix

A project manager can see their own project

Assigning someone to a project and marking them as managing it did not actually let them see the project. They saw only their own hours on it, and pulling a client report for it returned only their own entries — so the only way to give someone the reporting they needed was to make them a Manager across the whole account, which shows them every project you have.

Marking someone as managing a project now does what it says. They can see everyone's hours on that project, and report on it. Nothing else changes: they still see only their own time on projects they do not manage, and being assigned to a project is not the same as managing it.

Rates and costs are unchanged. Managing a project lets someone see the hours on it, not what those hours cost or earn. Internal cost rates stay with administrators, as they were.

If you raised someone's role as a workaround for this, you can set it back.

Fix

A Harvest import no longer gives up on one bad response

A large import walks through your Harvest data a page at a time, and if Harvest returned a server error on any one of those pages, the whole import stopped and started again from the beginning. On a big backfill that could throw away several minutes of work for a problem that had already cleared.

Those requests now wait a moment and ask again, up to five times, before giving up. Nothing changes for errors that are actually about your account — a wrong credential or a project that no longer exists still fails immediately, because asking again would only get the same answer more slowly.

We also fixed the email. An import gets three goes, and you were emailed "your import failed" after each of them — so an import that succeeded on its second attempt still sent you one alarming message first. You now hear from us once, after the last attempt, and only if the import genuinely did not finish.

Fix

Magic fill understands the company you named

Magic fill turns a sentence into an entry. Say "3h design for Acme yesterday" and it should fill in the duration, the day, the project and the task.

It was only ever shown your project names, never your client names. So when you named the company — which is how most people describe their work — it had nothing to match against. The hours, the task and the notes filled in correctly and the project was left on whatever was already there.

Where the same project name appears under more than one client, which is normal for a "Retainer" or a "Monthly", it was worse than not matching: there was no way to tell those projects apart, so one was chosen and the entry landed on the wrong company while everything else about it looked right.

It now matches on the client name as well as the project name. Naming the project still wins where both would fit, since that is the more specific thing to have said.

Worth a glance at recent entries if you use Magic fill on projects that share a name across clients.

Fix

A scheduled article now publishes itself

Articles carry a date, and one dated in the future is held back until that day arrives. That is what it looked like. In fact the blog index and the sitemap were built once and served unchanged, so a scheduled article appeared whenever somebody next deployed — which might be the right day, or a week later.

Both now refresh hourly, so an article appears on its own date without anybody doing anything. Nothing about how articles are written or dated has changed.

There is also a new Articles panel in the platform area, beside the what's new digest. It lists every article with its date, whether it is published or still scheduled, and when the next one is due — so it is possible to tell at a glance what is about to go out, and to read it first.

Fix

An entry that said "sending to Jira" is now actually being sent

The hourly check that compares your timesheet against Jira has a case for entries where your copy is the authoritative one and the Jira worklog has not been touched: send ours across. It marked the entry Sending to Jira and then did not queue anything, so the entry sat in that state indefinitely — reporting work in progress that nothing was doing.

Nothing was lost and nothing in Jira was wrong; the entry simply stopped describing reality, and because the broken version looked busy rather than failed, it raised nothing. Two entries in our own staging workspace had been stuck that way since 11 September.

The check now queues the sync first and marks the entry only once that has succeeded — and if the queue cannot be reached, it leaves the entry alone rather than promising a sync that will not happen. It shares the same grouping window as saving an entry does, so a save and an hourly check landing together produce one sync rather than two.

Fix

Saving a project no longer resets a setting you never touched

The project edit form loads its fields from several places at once — the headline details, the billing and invoicing fields, and the budget each arrive separately. If you saved while one of those was still outstanding, the form sent what it was showing, and what it was showing was a blank.

The catch is that a blank is a legitimate answer. An unticked checkbox means "no" and an empty number means "none", so there was nothing to warn you about and nothing in the saved project to suggest anything had gone missing. The case that brought it to light was Push linked entries to Jira as worklogs, which is on by default for new projects: the box rendered unticked, there was no reason to tick something that should already have been on, and saving turned it off.

The form now waits until every field can show what is actually stored before it will render, and says so plainly if a project cannot be loaded at all rather than offering you a form full of defaults.

Worth a look if you use the Jira integration. If worklog sync is off on a project you expected it to be on for, it is on the project's edit page under the Jira heading — tick it and save. Nothing else about the project was affected, and no time entries were changed.

New

An entry changed in Jira can now be settled, instead of staying stuck

If somebody edits a worklog in Jira after we wrote it, the entry here is marked Changed in Jira rather than quietly overwritten. That part was deliberate and has not changed: an edit made in Jira is somebody saying something, and discarding it silently would be the worst thing this integration could do.

What was missing was the other half. The entry had no way out. There was no action on it, and editing it here did not help either — the same check refused the change again, so the hours could move in neither direction. The flag was accurate and permanent.

Now a flagged entry shows you what actually differs — the hours and the day on each side, and who made the change in Jira — and offers two choices, taking neither on its own:

  • Use Jira's version — the hours and the day become what Jira holds. Nothing is sent back; the two sides already agree.
  • Keep mine — the worklog in Jira is rewritten to match this entry.

Often the answer will be that nothing you record has changed at all. An entry is flagged whenever the worklog is touched in Jira, and that includes somebody adding a comment — so the entry now says when the hours and the date are the same on both sides, rather than leaving you to guess what you are accepting.

Whichever you pick, the entry stops being flagged and normal syncing resumes.

An approved or invoiced entry still cannot be settled either way, because both choices would change a record this workspace has closed. Reopen the week first if the hours are genuinely wrong.

Fix

Tasks are in alphabetical order, and stored amounts stop growing decimal places

Tasks are listed alphabetically. They were sorted by an internal position that nothing in the app ever set, so almost every task shared the same sort key and the rest of the order came down to whatever the database returned first. In a task picker that means scanning a list to find a name that is sitting right there. This applies everywhere tasks are listed, including the picker on a time entry and the workspace Tasks page.

Amounts read back as you typed them. Money and percentages are stored to four decimal places, and the edit forms were showing all of them: a cost budget of 20,000 came back as 20000.0000, and a 20% tax rate as 20.0000. Nothing was wrong with the stored figure, but a form that hands back a number you did not type invites you to wonder what else it changed.

Now 20000 stays 20000, 1234.50 keeps its pennies, and a rate of 7.5% shows as 7.5. Anything genuinely more precise is left exactly as it is, so saving a form never quietly rounds a figure you set deliberately. This affects project rates and budgets, client defaults, and invoice settings.

New

A worklog changed in Jira now says so here, instead of nothing happening

Time recorded here flows to Jira. Until now, nothing came back the other way: if a colleague corrected your worklog in Jira, or deleted it outright, the entry in Time Trakkr carried on as though nothing had happened, and the two quietly disagreed for good.

Time Trakkr now checks Jira's own record of what changed, once an hour, and says what it found.

An edit in Jira flags the entry, it does not overwrite it. The entry is marked Changed in Jira and your hours are left exactly as you recorded them. This is deliberate. Somebody editing a worklog in Jira has not asked us to rewrite your timesheet, and an integration that silently changed hours you had already reported would be worse than one that stayed quiet — you would have no way of knowing a number had moved. The flag is the point: it tells you the two sides disagree and lets you decide which is right.

A worklog deleted in Jira unlinks the entry and keeps the time. The Jira link is cleared and the entry says why, but the hours stay. Deleting recorded time because Jira lost a row is not a decision an integration gets to make, and the hours may well have been real.

An entry that is approved or invoiced is never touched. It becomes a conflict with an explanation instead. Those hours have been signed off or billed, and changing them behind the approval is the one outcome nobody wants.

Nothing here is retroactive — the check looks forward from the point it was switched on, not back over every worklog that ever existed.

New

Export the detail a client report is actually made of

Exporting a project gave you a task summary: one row per task, with billable, hours and cost. Everything a client report is assembled from was missing — the date the work happened, who did it, what they wrote, and the ticket it was against. So the report that goes to clients was being rebuilt by hand.

The detailed time report now has a Client report export with the eight columns those reports use: Date, Project Code, Task, Notes, Hours, First Name, Last Name, External Reference URL. One row per time entry — two hours logged on the same task on the same day stay two rows — followed by a clearly marked TOTAL.

Costs and rates are deliberately not in it.

Project hours are now the way in. The headline total, and every task and person figure, opens the detailed report filtered to exactly what you clicked and the period you were looking at.

The report itself carries more. Project code, the ticket link, and the person's role on the project have been added; the existing export keeps billable, approval and invoiced state for internal use, and the project task summary export is unchanged.

Names

Harvest sends people's first and last names separately, and the importer used to join them and keep only the joined form. It now keeps both, so the export can print a real surname.

This only fills in when the Harvest setup import is re-run — the parts were never stored before, and they cannot be recovered from a display name. Splitting one on a space would be a guess: "Maria José Costa Silva" is "Maria José" and "Costa Silva", which no rule about spaces produces. Until an import runs, the whole name appears under First Name and Last Name is left empty, rather than a surname being invented for a document going to a client.

Ticket links

Built from the Jira site you are connected to, plus the issue key already stored on the entry. Entries with no key, or an organization with no Jira connection, export an empty cell rather than a guessed link.

Improved

Log time on another day without leaving the dialog, and set names for client reports

Change date now works when creating an entry, not only when editing one. If you did the work on Monday, you can say so while logging it. The workaround was to save it on today, then edit it and move it — two steps, and an entry briefly sitting on the wrong day.

The week lock applies exactly as it does everywhere else: a day inside a submitted week is refused, with the same message.

First and last name can be set on a person's profile. The client report export needs them in separate columns, and a profile only held a full name.

They are asked for rather than worked out. A full name cannot be split reliably — "Maria José Costa Silva" is "Maria José" and "Costa Silva", which neither the first space nor the last one produces — so the fields suggest a split as placeholder text and store nothing unless somebody types it. Left empty, the export uses the full name rather than inventing a surname.

People imported from Harvest already have theirs, since Harvest sends the two parts separately.

Improved

Jira mapping is rules now, and you can check one before you trust it

The admin Jira mapping used to be one grid of dropdowns: each Jira project pinned to one workspace project. It could say exactly one thing, and it said it about every issue in the project.

It is rules now. A rule matches on a whole Jira project, one issue type within a project, or a custom field value — so "bugs in ACME go to the Support retainer" and "anything where Team is Platform goes to the platform project" are both things you can now say. A rule can name the task as well as the project.

The most specific rule always wins, whatever order you added them in: a custom-field rule beats an issue-type rule, which beats a whole-project rule. A custom-field rule describes one issue and a project rule describes thousands, so letting the general one win would make the specific one impossible to express. Two rules of the same kind are separated by an order number you set when you add the rule, lowest first — it shows on each rule as #300. It only breaks ties within a kind: a whole-project rule set to #1 still loses to an issue-type rule set to #900, because specificity is the point and an order number should not be able to quietly undo it.

Leaving a rule's project as Ask every time is a real choice, not a blank one. That rule still wins and the picker asks — which is what you want where the answer genuinely varies and a vaguer rule would otherwise answer confidently and wrongly.

Check an issue before you rely on a rule

Type an issue key into Check an issue and you get the rule that wins, the rules that also matched and lost, and the project and task the time would land on.

This is the part worth using. Getting precedence wrong is silent: the hours go to the wrong project and look exactly like somebody picking the wrong project by hand. Seeing the rule you expected to win listed as beaten is usually the whole answer.

Two things that used to go quietly wrong

A mapping to an archived project used to be returned as though it were fine. Nothing validated the target, so a project somebody archived months ago kept being handed back as the authoritative answer, and starting a timer against it either failed for a reason the picker never mentioned or landed hours on a closed project. Rules are checked when you save them, and the picker now says when a rule needs an admin's attention instead of quietly guessing something else.

A rule keys on the issue type's id, not its name. Renaming "Bug" in Jira no longer silently stops a rule matching.

Existing mappings were carried over as whole-project rules automatically. One that pointed at a project that no longer exists was dropped rather than migrated — a rule pointing at nothing is worse than no rule, because it looks configured.

The assistant can read the rules, check an issue, and add a rule for you: "which rule sends ACME-142 to Support?" or "make bugs in ACME go to the Support retainer".

Improved

The Jira panel shows the issue's estimate

The Time Trakkr panel on a Jira issue used to show only what you had tracked. It now shows what Jira itself knows: logged in Jira, the original estimate, the remaining estimate, and how far through the estimate the work is.

Three deliberate choices in how those numbers are presented:

  • Jira's total and your unsynced time are never added together. Jira works out its "logged" figure from worklogs, so it already contains everything Time Trakkr has sent. Adding your tracked total on top would double it. Time recorded here that has not reached Jira yet is shown on its own line instead, so you can see the gap rather than having it hidden inside a total.
  • A missing estimate says "Not set". Not zero, and no progress bar. An issue nobody estimated is unestimated, not an issue with no budget that everything overruns.
  • Durations are Jira's own wording. "2d" means different things on different Jira sites, depending on how long the site says a working day is — so the panel repeats what Jira says rather than working it out.
Improved

Time reaches Jira exactly once, and the timesheet says whether it did

Worklog sync had three faults, all of them quiet.

Entries could be duplicated. If Jira accepted a worklog but the reply was lost on the way back, Time Trakkr never learned the worklog existed — and the next save created a second one. The issue then showed twice the time, and nothing anywhere said so. Every worklog Time Trakkr creates is now marked, so a retry finds what it already sent instead of sending it again.

Edits made in Jira could be overwritten. If somebody corrected a worklog in Jira and the entry was later edited here, our version simply replaced theirs. Sync now checks whether the worklog changed in Jira first, and stops rather than discarding somebody's correction. Approved and invoiced entries are never altered to match an upstream change either.

Failures were invisible. A sync that failed left no trace you could see: the entry saved, which was true, and nothing said Jira had not received it.

Every entry headed for Jira now says on the timesheet whether Jira got it — In Jira, Sending to Jira, Not in Jira, Changed in Jira, or Jira not connected — with Jira's own reason on hover, such as "you do not have permission to log work on this issue". Entries that were never going to Jira say nothing at all, so the status only appears where it means something.

A failed entry has a Retry. Two states deliberately do not:

  • Changed in Jira means somebody edited the worklog in Jira after we wrote it. Re-sending would overwrite their change, which is the thing we stopped to avoid — so open the issue and decide which version is right. Editing the entry here syncs it again.
  • Jira not connected is not a failure of the entry, and retrying fails identically. Reconnecting is the fix, and only that person can do it, so the link appears on your own timesheet rather than a colleague's.

Worklog push is on by default for new projects. It used to be off, and the cost landed on whoever set the integration up: the mapping rule resolved, the issue linked, the hours recorded — and nothing reached Jira, because a separate per-project switch nobody had mentioned was off. Linking an entry to a Jira issue is somebody saying what they want.

Existing projects keep their current setting. Turning it on for one is a deliberate choice, because it starts writing worklogs into your Jira from then on. Workspaces that do not use Jira see nothing either way — an entry with no Jira issue is never considered for syncing.

One date bug went with it: worklogs were sent as midday UTC, which is the following day in New Zealand and anywhere else past UTC+12. Time logged on a Wednesday arrived in Jira on Thursday. It now uses midday in your own timezone, worked out for the day the time was logged, so a backdated entry across a daylight-saving change lands on the right day too.

The remaining estimate

There is a new workspace setting for whether logging time reduces the issue's remaining estimate. It is off by default: reducing an estimate somebody else set is a change to their planning, not ours to make uninvited. Turn it on and Jira's burn-down starts tracking reality.

Fix

A monthly budget is measured against one month

If your project's budget resets every month, the figure on the project page was wrong, and it got worse the longer the project ran.

A 20 hour monthly retainer three months old reported "-43.08 remaining (0%)". The allowance was one month's; the spend charged against it was every hour the project had ever had. A project comfortably inside its budget looked catastrophically over.

It now counts only the month you are looking at. The project page opens on the current month, with arrows either side and Return to this month once you have left it. Stepping back to a month that genuinely ran over shows that month over, without affecting the one you are in.

Nothing about your budgets changed — no setting, no number, no alert threshold. What changed is the arithmetic underneath the display. If you have been ignoring a red budget figure on a retainer, it is worth another look; and if you had raised a budget to stop it reading negative, you can put it back.

Projects on a fixed budget are unaffected: their budget is one running total for the life of the project, and they still open on all time.

On reports, a monthly budget is still per month. The Budget vs actual table and the project-budget API compare a budget against whatever date range you asked for, and narrowing that range would answer a different question than the one you asked. So instead those rows now say budget is per month, and the API returns budget_resets_monthly beside them — a 20 hour retainer over a quarter shows 20 against 60, which is a project exactly on plan. For a like-for-like figure, ask about a single calendar month, or read the project page, which does the windowing for you.

The assistant knows this too, and will no longer tell you a retainer is over budget because you asked about a quarter.

Fix

The new entry dialog offers what you logged last, not what sits on the newest day

A new entry suggests the project and task you used last. It was picking the entry on the latest dated day rather than the one you most recently submitted — so once something sat on the newest day, anything you logged afterwards with an earlier date changed nothing and the form kept offering the same pair.

It now follows what you actually entered last, whatever day you dated it for.

More noticeable since backdating a new entry became a single click in the dialog: log Monday's work on Wednesday and the next entry should start from Monday's project, not from whatever was on Wednesday.

New

Click any hours figure on a project to see the entries behind it

The numbers on a project page could not be inspected. If a task showed 61.58 hours you could see the total and nothing else — who logged it, what it was logged against, or which entries made it up meant exporting a CSV and opening it somewhere else.

Every hours figure on the Tasks and Team tabs is now a link into the detailed time report, filtered to that project, that task or person, and the range you were looking at. Clicking a figure while viewing August opens August, not a default window.

"No task" works too, which was the point. It has no id and is often the largest row on the page, so a filter that could not name it would have dropped you on an unfiltered report at exactly the moment you asked about it.

The report says what it has been narrowed to and offers a single click to widen it again, so a drill-down is a starting point rather than a dead end. It also now shows the ticket an entry was logged against, alongside the notes.

Nothing changed about who can see what: the report already scoped members to their own entries, and it still does — a link naming somebody else is narrowed by the server regardless of what it asks for.

Improved

The project hours chart has a scale, dates, and your budget on it

The chart at the top of a project page showed bars and nothing else. There was no scale, so a bar's height only told you how it compared to the busiest week. There was no way to tell which week a bar was. And on a project with a budget, nothing showed where the budget was.

It now has:

An hours axis. Bars are drawn against real hours with labelled gridlines, so you can read a value off the chart and compare one project to another.

Dates under the bars. Labels thin out as the range grows, but the first and last are always shown, so the span the chart covers is legible at a glance.

Details on hover. Point at a bar — or tab to it — and the line beneath reads the period and its exact hours. Bars are keyboard reachable, so the numbers are not mouse-only.

A weekly or monthly view. Monthly is the default on projects with a monthly budget, because that is the period the budget is set in.

A budget line. Projects whose budget resets each month get a dashed threshold across the monthly view, and any month over it is drawn in red. The hover detail turns into hours spent and hours remaining against that month's budget.

The month you are looking at is marked, so you can see which bar the tables below are describing — the chart still shows every month, because comparing them is the point of having one.

The current month is drawn striped and labelled "so far", since a month that is still filling up is not comparable to a finished one and a solid bar invites exactly that comparison.

Weeks and months with no time now appear as gaps rather than being skipped. A project that paused for two months used to draw its bars side by side, which read as continuous work.

One thing this does not do: a project with a fixed total budget, rather than one that resets monthly, still has no line. A single total does not divide into a month, and drawing it as though it did would invent a number nobody set.

New

Change the date on a timer, including one that is already running

The timer popover used to say only "Running since 2:14 PM". It never said which day the clock was banking onto — so a timer on the wrong date looked exactly like a timer on the right one, and you found out later, on the timesheet.

It also always booked to today. Picking up yesterday's work meant starting the timer on today, stopping it, and then moving the entry.

Both are fixed, in the place you are already looking:

  • Before you start, the popover names the day and offers Change date, so work continued from another day can start on that day.
  • While it runs, it names the day it is logging to, and the same Change date moves it. The clock keeps running on the same entry — nothing stops, nothing is duplicated, and no time is lost.

The rules do not change. A move into or out of a week that has been submitted for approval is still refused, in both directions, and an approved or invoiced entry still cannot be touched. Moving a running timer out of a submitted week would quietly reduce a total somebody had already signed off, which is the same problem as moving time into one.

The assistant can see this too: asking about your timer now tells you the day it is logging to, not just how long it has been going.

Changing the date of a new or existing entry from the entry dialog shipped separately, and is unchanged.

Improved

See what you are approving, and land on the right person's timesheet

Three things the team asked for, all of them about getting to the right screen without a detour.

Approvals shows the entries behind the total. Each row now has a View button that opens that member's entries for the week, grouped by day and sorted by project. Until now the row gave you hours and an Approve button, and checking what those hours were actually spent on meant leaving for the timesheet, switching the member selector, and finding your way back to where you were.

It is available for approved and rejected weeks too, not only submitted ones — "what was in the week I approved on Tuesday" gets asked at least as often as "what am I approving now". The dialog is read-only: approve, reject and reopen stay on the row behind it.

The Team page opens the right timesheet. The Timesheet button beside a person now opens their timesheet. It used to open your own — a button sitting next to somebody's name that quietly showed you somebody else's week. Who you are allowed to view has not changed, and the selector at the top still switches freely once you are there.

The date arrows stay put. Stepping through days moved the ‹ › arrows, because the pill was only as wide as the date it was showing and "Today Wednesday, Sep 10" is a lot wider than "Mon, Sep 8". If you clicked twice quickly the arrow slid out from under the cursor. The pill now reserves room for the longest date it could show, so it stays exactly where it is.

New

Move an entry to another day, and choose which day a restarted timer lands on

Two things the timesheet could not do, both asked for by people moving over from Harvest.

Change date

Open an entry to edit it and there is a Change date control beside Delete. Pick a day and saving moves the entry there, with its hours, notes and project intact. Previously the only route was deleting the entry and typing it again on the right day, which loses whatever else was on it.

Restarting a timer on a day that is not today

Pressing Start on a row from an earlier day used to restart that entry's timer silently — and the time you then tracked was banked against that older day, not today. That is sometimes exactly right, if you are continuing the same piece of work. It is just as often not, and nothing said which was happening.

Start on an older row now asks:

  • Start timer on today's date — a new entry on today, carrying the same project, task and notes.
  • Yes, restart this timer — the old behavior, kept, because continuing yesterday's work on yesterday's entry is a real thing to want.

A row from today behaves exactly as before, with no prompt.

And a rule that was missing from two places

Submitting a week for approval freezes it. That was enforced when you created an entry, started a timer, or copied a day — but not when you restarted a timer on a row inside a frozen week, and not when you changed an entry's date. Both now refuse, with the same message, and moving an entry is refused in both directions: into a submitted week, and out of one. Editing the notes or hours of an entry already in a submitted week is unchanged.

The assistant can do both of these too — ask it to move an entry to another day, or to start a timer against a particular date.

Improved

Change how long a meeting ran before confirming its draft

Meetings rarely last exactly as long as they were booked for. Until now a calendar draft could only be confirmed at its scheduled length, so correcting it meant confirming the entry, finding it on the timesheet, and editing it there.

Each draft now has an hours field you can change before confirming. It starts at what the calendar said — shown alongside it, so you can see what you are overriding — and whatever you set is what gets logged.

Assigning a project to several drafts at once leaves any hours you have already typed alone.

Fix

A dismissed calendar draft comes back if the meeting is rescheduled

Dismissing a calendar draft was permanent in a way nobody chose. If you dismissed a call because it was about to be rescheduled, the rescheduled call never appeared — the day and time had changed, but the meeting was gone from your drafts for good.

A rescheduled meeting keeps the same calendar id, so once a draft for it had been dismissed, every later attempt to draft it was treated as a duplicate and quietly skipped.

A dismissed draft now comes back when the meeting actually moves — a different day, or a different time. It reappears under the new day, with the new length.

Dismissing still means dismissed for a meeting that has not changed: an unchanged call stays gone, which is the whole point of the button. And a draft you already confirmed is never re-offered, because the time is logged — if that meeting later moves, edit the entry rather than logging it twice.

New

Start a Time Trakkr timer from inside a Jira issue

Install the Time Trakkr app on your Jira site and every issue gets a Time Trakkr panel. Connect your account once, then start a timer without leaving the ticket. The entry carries the issue key from the moment it starts, so the worklog reaches Jira when you stop.

One box, either behavior. Leave the duration empty and Start timer runs a timer. Type 1:30 or 1.5 and the button becomes Log time, banking that much against the ticket without a timer at all — for the work you finished before you thought to start one. The panel also shows how much you have already tracked to the issue.

It fills in what it can and asks for the rest. Time Trakkr guesses the project and task from your admin's Jira project map, from time you have already logged against that Jira project, or from the issue's own project and type — and tells you which. When it cannot guess, the pickers are still there. That is the design rather than a fallback: somebody using this for the first time has no history to infer from, and refusing to work until a match is perfect would fail exactly those people. The issue title becomes the note, so the entry reads properly on a timesheet without anyone typing.

Projects are listed as client, code and name — "Northwind — [NW] Phase 1" — because a bare project name is not something you can choose by when two clients each have a "Phase 1". Archived projects are left out.

Connecting is a real authorization, not a pasted key. You approve it on Time Trakkr's own consent screen, the connection acts as you and under your own role, and it reaches only time entries, timers, projects, tasks, reports and Jira issue matching — never invoices, rates or the team directory.

Taking access back is one-directional today. Revoking under Developer → Connected apps stops it immediately, and the panel then asks you to connect again. Disconnecting the app on the Jira side does not yet revoke it here — it lapses on its own within thirty days. Use Connected apps if you need it gone now.

Stopping a timer you left running overnight still asks how long you actually worked, exactly as it does in the app. The panel shows that question rather than a generic failure.

Improved

Apps can connect with a client secret, and the consent screen says more

Connecting an app to Time Trakkr has always used PKCE — the app proves it is the same one that started the sign-in, without holding a shared secret. That works for anything running in a browser or on your own machine, and it is what the Claude connector uses.

Some integrations cannot work that way. An app hosted by another vendor holds its credentials on their servers and authenticates with a client secret instead. Time Trakkr now accepts either — but never neither, and which one an app owes is fixed when it is registered rather than decided per request. An app registered for PKCE cannot skip it by leaving the challenge out.

Client secrets are not self-service. Anyone can still register a PKCE app automatically. A secret is issued out of band, by a person, because an open endpoint handing out secrets would let anyone opt themselves out of PKCE.

The approval screen now mentions Jira. A grant could already match Jira issues to your projects to pre-fill a timer; the screen did not say so. It does now. Nothing about what a grant reaches has changed — only the description, which should have named it from the start.

Refreshing a connection now re-checks the secret. Previously a refresh token alone was enough for its full thirty days. For apps that use a secret, the secret is checked every time.

Not yet built: disconnecting an integration from its side does not revoke the grant here. Until Time Trakkr has a revocation endpoint, take access back under Developer → Connected apps, which does end it immediately.

New

Share an assistant conversation with a teammate

Until now a conversation with the assistant was yours alone. If you asked it something that turned out to matter to somebody else — why a project ran over, what is still uninvoiced for a client — the only way to pass it on was a screenshot or an explanation from memory.

Open Assistant, click ↗ beside a conversation, and pick a teammate. They get a link, and it appears in their Shared with me list.

What they get is read-only. They can read the transcript and see which tools the assistant used, but they cannot add to it, and nothing runs on their behalf when they open it. If they want to keep going, Start my own chat from here makes them a private copy — anything they ask after that runs with their access, not yours, and you never see it.

Sharing shows them the whole conversation. A transcript is a record of what the assistant found when you asked, using your access. If it looked up something your teammate cannot normally open, sharing the conversation shows it to them anyway. That is the same as pasting the conversation into a chat message, and it is a decision worth making deliberately — the dialog says so each time, and every share and removal is recorded in your audit log.

You can share only with active members of your organization. There is no public link. Remove someone's access at any time from the same dialog; it also ends by itself if you delete the conversation, or if either of you leaves the team.

The assistant will not share on your behalf. Ask it to, and it hands you a link to the dialog instead — it is reachable from Slack and from external AI clients where that warning cannot be shown, so a person confirms every share.

Sharing saved reports is next, and works differently: a report will re-run under the permissions of whoever opens it, so it can never show them more than they are allowed to see.

New

Share a saved report with a teammate

Open Saved reports, select Share on a card, and pick a teammate. It appears in their own Shared with me section, and they get an email with a link.

What travels is the definition, not your numbers. The filters, date range and grouping go across; the results do not. When your teammate opens it, the report runs against whatever they are allowed to see. Their totals may be smaller than yours and can never be larger — sharing a report cannot show somebody data they are not otherwise entitled to. A note at the top says so, because two people looking at "the same report" and seeing different totals is otherwise alarming.

One case is called out rather than left to guess at. If a report is filtered to a specific person and you share it with someone who cannot see that person's time, it would run and return nothing — which reads as "they logged no hours" rather than "you cannot see this". The report now says which it is.

The filters themselves are visible: the report's name, and the project, client or person it is filtered to. Sharing does not reveal the data behind them, but it does reveal that they exist.

You can ask the assistant to do this — "share my Northwind margins report with Priya" — and it will, because a report share cannot widen anyone's access. Sharing an assistant conversation still needs you to confirm it yourself.

Saved reports of every kind now reopen with their filters applied. Until now only custom reports restored their configuration; the Time, Detailed time, Invoicing and Detailed expense kinds opened on their default range with no sign that anything had been dropped. That was easy to miss when the only person opening a saved report was the one who saved it. It is not something to hand to a colleague, so it is fixed.

Fix

A Jira issue can be linked from more than one entry

Time entries carry two different kinds of reference, and until now they shared one field.

One is where an entry came from — its id in Harvest, Toggl, Clockify, or an external_reference sent through the API. That has to be unique, because it is what lets you re-run an import without creating everything twice.

The other is the Jira issue you are working on. Several entries pointing at one ticket is the entire point of that link — different people, different days, same piece of work.

Sharing a field meant the rule for the first applied to the second: an issue could be linked once per workspace, ever. The second person to log against a ticket saw a failure with no explanation, and their entry was not saved.

Jira links now have their own field. Link an issue from as many entries as you like. Import provenance keeps its uniqueness, so re-importing is unchanged.

Existing links were moved across, not lost. Editing an entry also used to clear its Jira link without saying so — that is fixed too.

Improved

Billable is set by the task, not on each entry

Each task has a Billable checkbox on its project. Until now nothing read it when time was logged: every new entry defaulted to billable, and the only way to record non-billable time was for the person tracking it to untick a box on their own entry.

So a task marked non-billable still produced billable hours unless everybody remembered, every time. Two people logging the same task could disagree, and the disagreement only showed up on an invoice.

The task now decides. Time logged against a non-billable task is recorded non-billable, everywhere — the timesheet, timers, the API and the assistant. The entry form shows what will be recorded and no longer offers a checkbox, because the answer is not a per-entry choice.

A non-billable project still overrides its tasks: its work is never billed, so a billable task inside it cannot make an hour billable.

To change whether work is billable, change the task — on the project's Tasks section. It applies to everyone at once, which is the point.

Existing entries are untouched. Their recorded value stands, so nothing that has already been invoiced or approved changes underneath you. Only new time follows the task.

Improved

Four fixes to the time entry dialog

All four came from somebody using this every day.

A click outside no longer discards your work. If you have typed anything, clicking the background does nothing — previously it closed the dialog and took the entry with it. Cancel and Escape still close it, because both are deliberate. A dialog you have not touched still closes on a click outside, since there is nothing to lose.

The same change fixes a subtler version of it: selecting text in the notes field and releasing the mouse just past the edge of the dialog used to count as a click outside. It now takes a press and a release on the background.

Cmd+Enter — or Ctrl+Enter — saves, from anywhere in the dialog. It starts the timer when you have not entered a duration, exactly as the button does. Plain Enter still adds a line to your notes.

The notes field grows as you type, up to about eight lines, then scrolls. It no longer opens as a one-line box you have to drag bigger to read what you wrote.

The projects you log against most are at the top, under a Recent heading, below anything you have pinned. Ordered by how many days you tracked against each over the last eight weeks, so it follows your work as it moves rather than ranking by whichever project had one long afternoon.

Improved

A new time entry opens on what you were last working on

The next thing you track is usually the last thing you tracked, and the empty form made you pick the same two values again every time.

Open a new time entry now and the project and task from your most recent entry are already there. Change either one as usual — nothing is saved until you save it or start the timer.

A few things it deliberately does not do:

  • It never fills in a project you have already chosen. If you used Magic fill, or picked something before the form finished loading, that stands.
  • It leaves an entry you are editing alone. Editing shows the entry's own project and task, as it always did.
  • It skips a project or task that has since been archived, rather than offering a selection that would fail when you save it. If only the task is gone you still get the project.

If you have never logged time, the form opens empty as before.

Improved

Runaway timer alerts go to the person's manager

When a timer runs past the escalation threshold, Time Trakkr emailed every owner and admin — about everybody. In a workspace where most people have no manager assigned, that meant a handful of admins receiving alerts all day about colleagues they have no relationship to.

Mail about somebody you are not responsible for reads as the product leaking, and the reliable outcome is that people stop reading any of it — including the one that mattered.

The alert now goes to the person's manager. If nobody is assigned as their manager, nobody is emailed: the person tracking still gets their own nudge at the earlier threshold, which is the first and most useful warning, and "no manager assigned" is precisely the statement that nobody has been given the job of watching.

Managers and admins can still ask the assistant "is anyone still clocked in?" at any time, which shows the whole workspace.

If you want these alerts to reach you, assign yourself as the manager of the people you look after, under Team.

Fix

Dismissing What's new now sticks

The dialog shows at most five entries at a time. Closing it marked only the five it had shown — so when more than five were waiting, the next page load brought it back with the rest. From the outside, the button did not work.

Closing it now clears everything it was entitled to show, whether or not it fitted on screen. Anything published afterwards still reaches you: only what was already waiting at the moment you closed it is marked as seen.

It showed up on a day with several notable releases at once, which is exactly when an interruption you have already dismissed is least welcome.

Improved

Ask the assistant who has not logged time

The Coverage report shipped as a page you open. It is now also something you can just ask for: "who is missing days this week?"

The answer comes back gap-first — the people with missing days and which dates they missed — rather than the whole grid. The grid is built for scanning with your eyes; a question deserves a sentence.

It applies exactly the same rules as the page, because it runs the same code. Weekends, public holidays and approved time off are never counted as gaps, so "nobody is missing anything" means that genuinely.

The same tool works in the in-app assistant, the Slack command, and Claude through the connector.

It will not chase anybody. It reads; it cannot log time on someone's behalf or send them a reminder. And it needs to know who is asking, so a connection made with an organization API key cannot run it — reading the team's approved time off requires a signed-in person, and without that the report would count everybody on holiday as missing their days.

New

See who has not logged time, without opening every profile

Reports → Coverage answers the question people were checking profile by profile: over a set of dates, who has not logged time, and on which days.

Pick a range. You get a row per person and a column per day, with the days somebody was expected to work but logged nothing marked clearly. People with the most missing days are listed first, so the ones worth chasing are at the top rather than wherever the alphabet puts them.

It knows what is not a gap. Weekends, public holidays on that person's calendar, and approved time off are shown as themselves and never counted as missing. A report that flags every Saturday and everybody on holiday is one nobody trusts, and you would end up back at checking profiles one at a time.

Hours are shown wherever any were logged — including on a weekend, because someone who worked Saturday should not appear as though they did not. A short day shows its number rather than being flagged; "nothing at all" is the signal worth acting on.

Available to managers and administrators.

Improved

Search and filter when picking people

Choosing people from a list of a hundred and thirty, with no way to find one, was not workable. Every bulk action now has a search box — so does editing who is in a group, where it matches on name or email.

Select all applies to whatever you are looking at. With a search typed it means "everyone matching", and says how many that is.

There is also a new filter by current designation, which answers the case that prompted this: set everyone to employee except the people already marked contractor. Filter to Employees only, Select all, apply — the contractors are not in the list, so they cannot be changed by accident.

Selections survive a search, so you can build a set across several of them. If anybody you selected is hidden by the search you have typed, the screen says so rather than leaving you to discover it afterwards.

Group and designation filters still clear your selection, because those change which people are eligible rather than only which are on screen.

Improved

Pinned projects keep their client heading, and the pinned block is bounded

Pinning a project floats it to the top of the picker, which is the point. But it also pulled the project out of its client's group, leaving a flat list of bare project names. If three clients each have a "CRM" or a "Monthly Retainer", the pinned block gave you no way to tell which was which.

Pinned projects now sit under their client heading, between a ★ Pinned banner and an All projects banner — the same shape as the rest of the picker and the same shape the projects list uses. A client whose projects are partly pinned appears in both places, which is intended: the pinned copy is the shortcut, the other is where the rest of its work lives.

Both banners matter. The pinned block covers several clients, so a heading that only marks where it begins leaves you unable to tell whether the fourth client down is still pinned or the start of the main list.

The same on the projects list, which had the same gap. Wherever pinned items appear, they now open with ★ Pinned and close with All projects — and when you have nothing pinned, neither banner appears, because a heading over the only list there is tells you nothing.

Search works on them again, too. The picker matches on the project name or the heading above it. While pinned projects sat under "★ Pinned", their client's name was in neither — so typing a client name found that client's unpinned projects and quietly skipped the pinned ones, which are the projects you use most.

New

Weekly rollups for account and project managers

Anyone holding a role on a project can now get a weekly summary of the projects they look after: hours logged over the past week, how much of each budget has been used, and — first in the list — any project with no time logged at all.

The quiet projects lead deliberately. You already know what your busy projects did; you were in the meetings. The one nobody touched is the thing you could not have known, so it is called out at the top rather than left to be spotted in a table.

Set up per role, not per person. Under Settings → Rollups, subscribe a role — Account Manager, Project Manager, whatever you use — and pick a day and hour. Everybody holding that role on a project receives it, for the projects they hold it on. A new account manager is included automatically, and somebody who hands over an account stops receiving it. Nobody maintains a list.

Roles are set on a project, or across a whole client at once. If you have not assigned any yet, nothing is sent — the rollup has nobody to address.

It contains no money. Rates and margin are visible to a subset of roles on screen, and an email is forwardable in a way a screen is not, so revenue stays out of it by default rather than by accident.

Anyone receiving one can switch it off for themselves under their own notification settings, and every rollup carries the link to do it.

August 2026

Fix

A handful of fixes

Pinned projects keep their client. Pinning a project moved it to the top of the list and, in doing so, dropped the client heading that told it apart from every other project of the same name. Pins now keep the same shape as the rest of the list: the client, then its projects underneath.

Settings → Groups was a dead end. The page opened without the settings navigation, so the only way to any other settings screen was the browser's back button.

The Anthropic key field says what is wrong. Typing something too short left the Save button inert with no explanation, which is indistinguishable from broken. It now tells you what a key looks like and where to find it.

New

Roles on a project, and setting them a client at a time

You can now record what somebody does on a project, not just that they are on it. Open a project's Team panel and each person has a role — Account Manager, Project Manager, or anything else in your work-role list.

A role is not a permission. Naming somebody the account manager says in what capacity they are involved. It does not give them the project's rates or margin — that stays the separate "manages this project" tick, so the two decisions remain yours to make independently.

Setting them one project at a time would not be reasonable, so you can apply a role to every active project of a client in one action. An account manager is usually a fact about the client rather than about each project, and this turns hundreds of decisions into a few dozen. It is available in three places, because it is the sort of thing you think of in all of them:

  • Clients → a client → Team roles
  • Settings → Bulk actions → Assign a role across a client
  • A project's Team panel, for the client that project belongs to

All three do exactly the same thing, and all three appear in your bulk-action history. Existing rates and per-person budgets are never touched — only the role changes. People who were not yet on a project are added to it.

This is groundwork: roles are what upcoming scheduled rollups will be addressed to, so an account manager can get a weekly summary of their projects without anybody maintaining a mailing list.

Fix

Timesheet notifications go to the right person, or nobody

If somebody submitted a timesheet and had no manager assigned, Time Trakkr emailed every owner and administrator in the workspace. In a team where most people have no manager set, that meant everybody with an admin role received everybody else's updates.

It reached the point where one person asked whether the product was leaking other people's data, and said the noise had made them wary of connecting their calendar at all.

Now a submission notifies the submitter's assigned manager, and nobody if they have none. The week still appears on your Approvals screen either way — what stops is mail going to people with no connection to the person who submitted it.

You may notice quiet. If your team has not assigned managers, timesheet emails will stop entirely until you do. Set them under Team → a person → Reports to. Approvals themselves are unaffected.

Leave requests are deliberately unchanged and still notify administrators when nobody is assigned: an unnoticed timesheet waits harmlessly in a queue, while an unnoticed time-off request leaves somebody waiting on an answer.

Every one of these emails now carries a link to turn it off, and your notification settings have moved up the account menu next to Account & security.

New

Use your own Anthropic key

Time Trakkr's AI features have always run on our Anthropic account. An owner or admin can now add the workspace's own key under Settings → Assistant, and all three features that call a model will use it: the in-app assistant, natural-language time entry, and the morning calendar drafts.

Prompts then go to your Anthropic account, under your commercial terms and your retention settings. Usage is billed to you rather than to us.

The key is checked against Anthropic before it is saved, so a mistyped or revoked key is caught while you are still looking at the screen — not by somebody else, days later, in the middle of a question. Once stored it is encrypted and cannot be read back by anyone, including us. Settings shows the last four characters so you can tell which key is installed.

This is not the same as connecting Claude or ChatGPT over MCP. That already runs on your own subscription and is unaffected. This is about the assistant inside Time Trakkr.

Two things we have deliberately not done. We never fall back to our key when yours stops working — you chose your own account, and quietly moving your prompts back through ours is not a decision we should make for you, so the assistant says what went wrong and stops. And using your own key does not change what we store: your conversations stay in Time Trakkr, exactly as before.

If a key is rejected, we stop retrying and email every owner and admin, because the alternative is a settings page nobody has open. Time entry and calendar drafts keep working throughout — they fall back to simpler matching rather than failing.

Settings → Assistant is also where the switch to turn the assistant off for the whole workspace now lives.

Improved

Password, email and two-factor in one place

Changing your password meant finding your own row under Team, opening your profile, and choosing a Security tab — while the menu entry actually called Account security led to a page offering only two-factor. A reasonable place to look, and the wrong place to find nothing.

Your password, the address you sign in with, and two-factor authentication are now all on that one page: your account menu → Account & security.

The Security tab on your workspace profile now points there rather than offering a second copy. Your calendar connection stays where it was, because that genuinely is per-workspace.

These settings belong to your account rather than to a workspace — one identity across all of them — which is why they sit outside the workspace pages now.

Improved

The assistant can draw charts

Asked to chart something, the assistant used to draw blocks of text in the shape of a bar chart and suggest you paste the numbers into Excel. Now it draws the chart.

Ask for vacation hours by month, tracked time by person, or a project's hours across the quarter, and the answer comes back as a bar or line chart in the conversation, with the numbers beneath it.

It is deliberately restrained. A chart appears when a trend or a comparison is the point — several months, several people. Two numbers get a sentence, because a sentence is clearer. And the assistant still says the date range and the headline figure in words, since a picture on its own does not tell you what it covers.

Charts scale to the width of the conversation, so they work on a phone, and they carry a text description for screen readers.

Improved

Change the email you sign in with

There was no way to change the address you sign in with. Correcting a typo, or moving somebody from a personal address to a work one, meant asking us to edit the database.

Change your own under Team → your name → Security. We send a confirmation link to the new address, and the change happens when you open it — until then your current address keeps working, so a mistyped address costs you nothing.

An administrator can change yours from Settings → Team, for the case that matters most: an address you cannot receive email at, where waiting for you to click a link would be no help at all. That applies immediately, and a notice goes to both the old and new addresses so a change you did not ask for is visible to you.

Two things worth knowing. Your password and any linked Google account are unaffected either way. And if you belong to more than one workspace, an administrator cannot change your address — your account is one identity across all of them, and one workspace should not decide how you sign in to another. You can still change it yourself.

Imports are unaffected: people are matched by their id in Harvest, Toggl or Clockify, so changing an address will not create a duplicate.

Improved

Groups

You can now group people. Create a group under Settings → Groups, tick who belongs, and the tags appear beside each person on your Team list.

Somebody can be in as many groups as you like, or none. That is the difference from a department: a department says where a person sits in the organization and they have exactly one, while a group is a label you invent for whatever reason — a squad, a shift, a client pod, whoever is on call this month.

Settings → Bulk actions gains Add to a group and Remove from a group, and its people list can be filtered by group — so "everyone on x-team" is a way to select people for any bulk action, not only a way to tag them.

The Team page gains a group filter beside the department and status pickers, and the three combine — so you can ask for contractors on x-team who have not yet accepted an invitation.

The assistant knows about them, and about the other Team filters too. "Who is on x-team?", "Who is in Engineering?", "List the contractors" and "How many hours did x-team track last month?" all work — in the app, in Slack, and through the MCP connector. Managers and admins can ask about anyone; everyone else gets their own record, exactly as on the page.

Any administrator can create and edit groups. Everyone can see them.

Fix

Changing someone's email no longer confuses the next import

People brought in from Harvest, Toggl or Clockify used to be matched purely by email address. That worked until somebody's address changed — correct a personal address to a work one, run another import, and the import no longer recognized them. It added them a second time, sent their new time to the new record, and left their history on the old one.

Nothing warned you this would happen, which made an ordinary edit on a profile page quietly expensive.

Everyone imported now carries their id from the system they came from, and that is what the next import matches on. Email addresses can change on either side without consequence.

Nothing to do. People imported before this change are linked automatically the next time you import, matched once more by their old address and remembered by id from then on.

Fix

People no longer show as pending after they have joined

Somebody could appear on the Team page as a signed-in member and on the invitations page as still pending, at the same time. One person had two invitations issued a month after they joined.

Three things caused it, and all three are fixed.

Inviting an existing member is now refused. The bulk invite already skipped anyone who had joined; inviting a single address did not, so it quietly created an invitation that could never resolve.

Joining by email domain now closes any outstanding invitation. People who join because their address matches your workspace domain never open an invitation, so nothing ever marked one as answered.

The invitations page no longer lists invitations for people who are already in. Accepting one would change nothing, so it is not something to act on.

Nothing was lost and no invitation was revoked — the records remain, they are simply no longer presented as outstanding when they are not.

Improved

Profile pictures

People appeared everywhere as two grey initials. You can now upload a picture under Account & security, and it shows wherever you appear.

If you sign in with Google and never upload anything, your Google picture is used automatically — nothing to do. Upload one and it takes over; remove it and the Google picture comes back rather than dropping to initials.

PNG, JPEG or WebP, up to 2 MB.

Only you can set yours. There is no way for an administrator to upload a picture on somebody else's behalf, which is deliberate.

Fix

Address lookup works again — and the field always accepts typing

Adding someone's address had stopped suggesting anything, and on some domains the Address field could not be typed into at all. A note underneath advised entering the address by hand, which was the right advice and impossible to follow — the field it referred to was the one that had stopped accepting input. That is fixed.

Suggestions now come from Google's current Places service rather than the older one, which is what had quietly become unavailable. In normal use the difference is invisible: start typing, pick a suggestion, and City, Region, Postcode and Country fill themselves in. Postcodes work as a starting point too — typing SW1A 2AA finds the street.

The more important change is underneath. The lookup can no longer take the manual path away. Every field, including the first, is an ordinary text box that stays typeable whatever the suggestions are doing — loading, failing, or switched off entirely. If Google is unreachable, the list simply does not appear and you fill the form in yourself, exactly as you would have anyway. Addresses that Google does not know — rural, new-build, much of the world — were always going to need that, and now it is guaranteed rather than assumed.

Suggestions are restricted to actual addresses, so an address field no longer offers you coffee shops.

Improved

Ask the assistant about one person's time

You could already ask the assistant for a total grouped by person. You could not ask it about one person — there was no way to turn a name into something the time tools understood, so it fell back to telling you to open Reports and filter by hand.

Now you can ask directly. "When did Jordan first log time?", "How many hours has Sam tracked to Acme this quarter?", "What has Alex been working on lately?" — all answered in the conversation.

Two things worth knowing. Asking for the first entry and the latest one are different questions, and the assistant now knows which end of the range it is looking at, so a long history no longer hides the answer. And when there are more entries than it can read at once it says so, rather than summarizing the part it happened to see.

Who can ask about whom is unchanged. Managers, admins and owners can ask about anyone in the workspace, exactly as they can in Reports. Everyone else gets their own time and their own record only — asking about a colleague returns nothing rather than an error, and naming a colleague explicitly still returns your own hours.

Fix

Profile fixes — phone numbers, address suggestions, and your own employment record

Three fixes to the People profile form.

Emergency contact numbers were cut off. The row gave the phone the same width as Name and Relationship, but a phone is two controls — a country picker and the number — so the number was left with about 88 pixels, enough for +1 267 62 and nothing more. The number field now gets the space it needs, and if a window is ever too narrow for both, the number drops onto its own line rather than hiding digits.

Address suggestions appeared when you opened the page. Opening a profile that already had an address searched for the address it was showing and dropped a list of suggestions over the form — including, at the top, the one already saved. Suggestions now appear only in response to typing.

HR admins can edit their own job title, dates and employment note. These fields are records about someone rather than fields they maintain, so for most people they stay read-only and say who to ask. But an administrator — and an owner, who is one by default — manages their own record on the same terms as anyone else's. Previously they had to find themselves in the People list to do it, which was a detour with no rule behind it. The permission is unchanged: the server still requires HR administrator rights, and for everyone else these fields remain read-only.

Improved

One page for notifications, not two

There were two pages called Notifications. One under Settings, listing the emails Time Trakkr sends about the product. One on your own profile, listing reminders, approval notices and the timer warnings. Same label, different contents, reached from different menus — so somebody trying to stop an email had to guess which, and half the time guessed wrong.

They are now one page: Team → your name → Notifications. The operational notices are where they always were, and the product emails sit beneath them.

Two things are worth knowing about that second group. They follow you rather than a workspace, so if you belong to more than one, turning something off turns it off everywhere. And only you can change them — an administrator can set your reminders, because those are about running the workspace, but not these. They are consent to be written to, and that is yours.

The old Settings link still works and sends you to the right place. Unsubscribe links at the foot of those emails are unchanged.

Fix

Imported projects respect "automatically add to new projects"

Team → a person → Automatically add to new projects puts that person on every project as it is created. It was working for projects made in the app, duplicated from an existing one, or created through the API — but not for projects brought in by a CSV import, which is the one route by which a workspace creates a lot of projects at once. Anyone relying on the setting had to go back and add themselves by hand to exactly the batch where it would have saved the most work.

Importing now assigns them, in the same transaction as the import itself, so a part-finished import cannot leave some projects assigned and others not.

Importing from Harvest deliberately still does not do this. That import brings across Harvest's own record of who is on which project, and adding people on top of it would put them on projects Harvest says they are not on. The setting itself comes across from Harvest, so somebody who had "access to all future projects" there keeps it here.

Fix

Emails go to the people they are actually about

Three of our emails were being sent on the wrong basis, and the first dry run against real accounts made all three visible at once.

The 45-day notice was being sent to the wrong people. "The last one of these" is meant for an account nobody has touched in six weeks. It was going to anyone whose own timesheet had been empty that long — so a workspace being used every day could still have three people told, on the same morning, that nothing had been tracked there for over a year and this was the last they would hear from us. It now goes to the workspace owner, and only when nobody at all has tracked anything for 45 days.

The weekly check-in claimed something it had not measured. It said "nothing has been tracked in your workspace", while counting only the reader's own entries. It now says what it means: you haven't tracked anything for about this many days. It also stops once you are well past a quiet week — if you do not keep a timesheet at all, a nudge about an empty one is not useful.

The budgets tip was going to people who cannot set a budget. It even opened by saying "you are getting this because you are an admin", to plenty of readers who were not, above a button that would have refused them. Tips that ask somebody to change a workspace setting now go to the people who can.

Underneath, every lifecycle email now declares who it is about — one person, or the workspace — and the least role that can act on it. Copy, recipients and the figures quoted are all read from that one declaration, so they cannot drift apart again.

And you can turn any of it off. Settings → Notifications lists each kind of email with a switch, and turning something back on works there too. Previously the only control was the unsubscribe link in a footer, which could not be undone from inside the app — the confirmation page had been pointing at a Notifications page that did not exist. Invoices, invitations and password resets are unaffected; those are not marketing and have never been optional.

Improved

The what's-new email is monthly now, and reviewed before it goes

The what's-new email used to go out weekly, built from whatever release notes carried that week's dates. That had two problems.

A week of product work is often one fix and a copy change, which makes for a thin email — and a thin email teaches people to ignore the next one. It is monthly now, so there is usually something worth reading.

It also carried everything we shipped, including work that isn't product news at all. A blog and a set of comparison pages are real things we made, and neither belongs in the inbox of somebody paying us to track their time. Release notes can now say who a change is for, and only product changes go in the email.

The last change is the one that matters most: a person reads the month's draft before anybody else receives it. The notes are written by whoever shipped each change, which is the right place to record work and the wrong place to decide what belongs in a customer's inbox. If the draft is wrong, it gets fixed before it goes out — and if nobody reads it, nothing sends.

Nothing shipped in a month? Then no email. A digest that says "nothing this month" is one people filter, and then the month that mattered is filtered too.

Fix

Emails now have a proper dark mode

If you read email in dark mode, anything we sent you looked worse than it should have. The design told mail clients it supported dark mode and then gave them nothing to work with, so they inverted it themselves — the warm cream card came out a muddy grey, the small print at the bottom dropped to the point of being hard to read, and our logo sat on a hard white rectangle.

Every email now carries a real dark palette: a warm dark card rather than a blue-grey one, small print that stays legible, panel borders that hold their shape, and a link color lifted so it reads against the darker background. The logo sits on a deliberate rounded chip, which is honest about the fact that the file has no transparency.

There is a matching set of tests, so an email that gains a color without a dark counterpart now fails the build rather than being discovered in somebody's inbox.

Improved

Assigned projects are grouped by client

A project name does not have to be unique, and in practice it often isn't. If you run the same piece of work for many customers — a campaign, a retainer, an onboarding — you have one project per client, all with the same name. That is how it should be.

The Assigned projects tab on a person's profile did not show it that way. It listed project names and nothing else, so fifty-four separate clients' projects appeared as fifty-four identical rows. There was no way to tell which was which, and no safe way to remove one.

They are now grouped under their client, the way projects appear everywhere else in Time Trakkr, and sorted by client and then by project.

Nothing has changed about the assignments themselves — only how they are shown. If that screen looked like it contained duplicates, it did not: those were different projects for different clients all along.

Fix

A daily backup now runs at the same time every day

A daily backup used to drift. The check that decides whether a destination is due asked for a full 24 hours since the last run, but it only gets to ask once an hour — so a backup taken at 10am was always a few minutes short when 10am came round again, waited for 11am, and went later every day after that. Over a week a morning backup walked into the evening.

Two things caused it and both are fixed. A destination is now allowed to be due slightly early, so it settles on the time it first ran at, and the time recorded is when the export began rather than when it finished — so a slow backup no longer pushes tomorrow's later by however long it took.

You do not need to change anything. Existing destinations settle onto a consistent time from their next run.

There is still no way to choose the hour. If you need a backup inside a specific window, tell us — it is a small change, and knowing the window you need is the part we are missing.

New

Release notes, and a log of everything that has shipped

Until now the only way to find out something had changed was to notice it.

Two things are new:

  • A dialog after you sign in, showing changes from the last month that were significant enough to be worth interrupting you — at most five, newest first. Dismissing is remembered per entry, so a note you have already read will not come back, and a note published after you dismissed one still reaches you.
  • Settings → What's new, which lists everything, including the small changes the dialog deliberately leaves out. Each entry expands to the full detail.

The dialog stays rare on purpose. One that fires for every change is one people learn to dismiss without reading, and then the one that mattered gets dismissed too. A week of small fixes is usually no dialog at all.

You can also ask the assistant "what's new?" and it will answer from the same list. It only knows about what this deployment has actually released, so if it says a feature is not there yet, that is the truth rather than a gap in what it was told.

New

A workspace with no tasks can add the standard set

People log time against a task, so a workspace with none has nothing to record against — and until now the Tasks page said "None yet." twice and left it there.

New workspaces have always been given a starter set at creation. What was missing was any way back for a workspace that did not get one, or that cleared its list and wanted it again. The only fix was direct database access.

Now, when the task library is empty, Settings → Tasks offers to add the standard set: Project Management, Engineering, Design, QA & Testing and Meetings as common tasks, plus Research, DevOps, Documentation, Support, Business Development, Admin and Internal as optional ones.

They are ordinary tasks. Rename them, change which are billable, move them between common and optional, or delete the ones you do not want. "Common" means a task is copied onto every new project automatically.

Owners and admins see the button; everybody else is told who to ask. And it will not merge into a list that already has tasks in it — if you already have some, add what you need individually rather than having twelve appear alongside them.

New

Timers that run too long now tell you

A timer records nothing until it stops — so one left running on Friday afternoon becomes a 70-hour entry the moment somebody notices on Monday.

Two things now prevent that:

  • A nudge after 10 hours. The person who started the timer gets an email; if it is still running after 24 hours, workspace admins are told as well. Each is sent once per timer, so a timer running for a week does not produce a week of email.
  • Stop refuses an implausible duration. Above 16 hours, pressing Stop saves nothing and asks for the hours you actually worked. That number is written exactly as you type it and never re-checked against the clock. Cancel and the timer keeps running, so nothing is lost either way.

Nothing is ever stopped or rounded on your behalf. An admin can change both thresholds under Settings → Time policy.

New

Webhooks can send invoice and time-entry changes, not just new entries

Webhook endpoints could only ever subscribe to one event, time_entry.created, even though Time Trakkr was already raising five others internally. Those five went nowhere.

You can now pick any combination of:

  • time_entry.created, time_entry.updated, time_entry.deleted
  • invoice.created, invoice.sent, invoice.paid

Each endpoint has its own selection, on the Developer page, and receives only what it asked for. Existing endpoints are unchanged and keep receiving time_entry.created.

What has not changed: delivery is still attempted once and is not retried, and a failure is recorded in the deliveries list rather than raised to whoever triggered it. A webhook your receiver rejects will never block someone from logging time. If you need to be certain you have everything, treat the API as the record and webhooks as a prompt to go and read it.

New

Scheduled backups to storage you own

Your data can now leave on your own schedule, to storage you already control. Connect Google Drive, Dropbox, or Box under Settings → Backups, pick a cadence and how many archives to keep, and each run writes a ZIP of CSVs.

Every archive carries a manifest listing each table, its row count, and — just as importantly — what was withheld and why:

  • No personal HR data. Compensation, performance reviews, leave records, employee profiles and documents are scoped to the individual rather than the organization, and are never exported.
  • No credentials. The archive contains no OAuth tokens and no webhook secrets.
  • No internal machinery. Request-deduplication keys, delivery logs and unsaved drafts are your rows, but not records you would restore from.

A failed run emails you once, with the reason, rather than failing quietly.

Fix

Copying into a locked week is refused

Adding a time entry to a week you have already submitted has always been refused. Copying rows into one was not — the copy succeeded and inserted fresh open entries into an approved week.

Copy now refuses the same way, and the message names the week that is locked.

Using Time Trakkr already? The same notes are under Settings → What’s new, and the help center covers how each feature works.

The hours, the invoices, and what they earned

Free for unlimited people. One paid plan, $10 per member per month, when you want HR, scheduling, timesheet approvals, scheduled backups, email from your own domain and scheduled reports.

Start free