Time Trakkr made this page, and Time Trakkr is first on the list. That’s the bias, stated up front so you can price it in. Everything below it is ordered on merit, every figure links to the vendor whose price it is, and each entry says what that tool does better than ours — because a comparison where the author wins every row isn’t worth reading.
Two different things get called AI agent time tracking, and most pages blur them. One is the time a person spends working with an agent: you, in the terminal, while Claude Code runs. The other is the time the agent itself spends working, which can carry on while you are in a meeting, and several agents at once. A tool built for one tells you little about the other, so every entry below says which it records.
The second split is whether the tool can bill a client. Several of the best tools here measure carefully and invoice nothing: you export a timesheet and bill from somewhere else. If a client is going to pay for agent work, find out where that line on the invoice will come from before you choose.
Hybrid teams, people working alongside agents, have one more question: whether agent hours are added to a person’s hours. Added together, they inflate utilization and blur what anyone actually worked. Kept apart and still counted toward the project, they give you both numbers. We did not find enough tools built for hybrid teams to make a separate list, so this one covers both.
| Tool | Price | Free plan | Invoicing | Profitability | Approvals | Monitoring | Imports history | AI agents |
|---|---|---|---|---|---|---|---|---|
 Time Trakkr | Free; paid plan $10 per member per month | YesFree for unlimited people | YesIncluded | YesIncluded, not a premium tier | YesPaid plan, $10 per member per month | NoDeliberately not built. No screenshots. | YesHarvest, Toggl, Clockify tokens or CSV | YesMCP server; Claude Code and Codex time at agent rates |
 Timely | From $9/user/mo annually ($11 monthly) | NoTrial only | NoExports to invoicing tools | YesProject health and rates | YesPremium, $16/user | YesPrivate to the user by design | YesCSV import | NoNo MCP found; logs your own use of AI apps |
| KKeito | From $16/mo per workspace billed annually (Solo, one user); Pro $39/mo, unlimited users | No14-day trial | YesEvery plan | NoNot that we could find | NoNot that we could find | No | NoNot that we could find | YesClaude Code and Codex sessions; agent SDKs and MCP |
| DDevClocked | Free; Pro $16/mo, Ultra $26/mo, Business $29/seat/mo | YesNo agent tracking on Free | YesUltra, $26/mo | NoNot that we could find | YesBusiness, $29/seat/mo | YesEditor and browser activity | No | YesClaude Code, Codex and Cursor runs with token cost, Pro and up |
| HHourtick | Free; Pro $29/mo per workspace, no per-seat charge | YesUnlimited members, 500 MB a month | YesBoth plans | NoNot that we could find | No | No | No | YesChat agents’ time and cost on tasks; coding agents not verified |
| WWakaTime | Free; Premium $14/mo ($12.83 annually); Team $21/dev/mo | YesOne week of history | NoMeasures only | No | No | YesEditor activity, automatic | No | YesAI vs human code and AI spend, Premium and up |
| TTiming | From $10/mo (Professional, monthly); other plans not verified | No30-day trial | NoSends entries to billing tools | No | No | YesMac activity, private to the user | No | YesYour time in Claude Code and Codex, not the agent’s |
| cccusage | Free, open source | YesFree | No | No | No | No | No | YesToken use and cost per session; not time |
| aagent-hours | Free, MIT licensed | YesFree | NoCSV, JSON or Markdown to attach | No | No | No | No | YesHours from Claude Code and Codex logs, human and AI split |
| cclaude-code-timelog | Free, Apache-2.0 licensed | YesFree | NoTimesheet reports | No | No | No | No | YesClaude Code sessions and tickets, logged locally |
Prices verified 2026-08-09 from each vendor’s own pricing page — linked from the tool name so you can check rather than take our word for it. Per user per month unless noted. Vendors reprice without warning; if you find something out of date here, tell us and we’ll fix it.
AI agents: whether the vendor runs its own MCP server for assistants such as Claude, and whether it records the time AI coding agents spend working. Checked 2026-11-05 against each vendor’s own help and developer pages; community-built servers are not counted. “Not that we could find” means we looked and found nothing, not that it cannot exist.
The back office behind billable work: time, margin and invoicing free for unlimited people, with approvals on one paid plan, for teams that bill for their work.
Nothing — this is our product. We’re first on this list because we wrote it, and the rows below say where the others are ahead.
Automatic tracking: it records what you worked on and drafts the timesheet with AI, rather than asking you to start a timer.
Capturing time nobody remembered to log. The automatic memory of your day is the best answer in this list to the problem of hours reconstructed on a Friday afternoon, and we have nothing equivalent.
Time tracking and billing built around agents: agent sessions sit beside people’s hours on the same project and become invoice lines of their own.
Breadth and pricing shape. Besides Claude Code and Codex it connects agent frameworks through SDKs and a CLI, records token cost as an expense, and one flat workspace price covers unlimited people and agents on Pro and above.
Automatic tracking for developers: editor, terminal and browser activity plus coding-agent runs, mapped to repositories and pull requests.
Automatic capture and cost detail. It records a developer’s editor and browser activity without a timer, tracks Cursor as well as Claude Code and Codex, and shows token cost per session and model.
Tasks, team chat and time tracking in one place, where people hand work to chat agents and see what each one took.
Delegating to agents. You can mention an agent in a task or chat and see its time and cost there, and both plans include unlimited team members with invoicing at no fee of its own.
Coding activity metrics from editor plugins, now including how much of the code AI wrote and what the AI cost.
Insight into AI-assisted coding. Plugins for a very wide range of editors, AI-written versus human-written code, and AI spend per developer answer questions about engineering that billing tools do not ask.
Automatic time tracking for the Mac, which records the project and file a person works on, including in terminal sessions with Claude Code or Codex.
Recording your own time without trying. It captures what you worked on in the terminal while an agent runs, with no plugin and no timer, and keeps it private to you.
A command-line report of token use and cost by day, month and session, read from the logs coding agents keep on your machine.
Knowing what the tokens cost, for nothing. It reads local logs from Claude Code, Codex and other agent CLIs, needs no account, and is far more widely used than anything else on this list.
A local tool that works out billable hours afterwards from Claude Code and Codex session logs, split between human and agent time.
Retroactive hours. It needs nothing installed in advance, so it can produce an hours estimate for work already done, and it splits the time between you and the agent.
A Claude Code plugin that logs sessions, projects and tickets to local files and reports them as a timesheet.
Keeping everything on your machine. Sessions are logged locally as they happen, it can backfill from old transcripts, and it picks up ticket numbers so time lands against tickets.
Recording the time an AI agent spends working, so it can be reported, costed and, where a client is paying, billed. It is different from tracking the time a person spends using an AI tool, which is what most automatic trackers record. Both are useful; they answer different questions.
Yes, with a tool that records the agent’s working time and also invoices. Keito, DevClocked from its Ultra plan, and Time Trakkr all put agent work on client invoices. Time Trakkr bills it at an hourly agent rate you set, on lines labeled as agent time; DevClocked’s invoices show agent runs with their token cost.
Usually not. Added to the hours of the person who ran it, agent time can make someone running three agents look as if they worked a thirty-hour day. Keep the two apart and count agent time toward the project instead: its budget, its margin and its invoice. Time Trakkr keeps agent time out of utilization and in budgets and profitability; check how any other tool reports it before relying on its utilization figures.
Track people’s time as you do now, and add something that records agent sessions against the same projects. The test is whether both land in one place with separate totals: the person’s hours, the agent’s hours, and a project total that includes both. Keito and Time Trakkr record people and agents side by side, and DevClocked records both for each developer.
Yes. ccusage reports token use and cost from local logs, agent-hours turns Claude Code and Codex session logs into an hours estimate, and claude-code-timelog logs Claude Code sessions as they happen. All three are free and run on your own machine; none of them invoices.
No. Token cost is what the model usage cost; agent time is how long the agent worked. Clients are usually billed for the work rather than the tokens, which is why Time Trakkr bills agent time at an hourly agent rate and shows token cost only as an estimate that never reaches an invoice.
Switching from one of these? We import from Harvest, Toggl Track and Clockify with a token, and from anything else with a CSV — clients, projects, rates, people and every hour you have logged, with their original dates. Also worth reading: best time tracking software for agenciesbest time tracking software for freelancers.
Time, profitability and invoicing free, for one person or fifty. One paid plan when you want more.
Start free