Time Trakkr's AI features normally run on our Anthropic account. You can supply your own API key instead, and everything that calls a model will use it.
This is useful if you already have a commercial agreement with Anthropic, if your legal team has approved Anthropic but not us as a subprocessor, or if you would rather the usage was billed to you.
Three features in Time Trakkr call Anthropic, and all three switch to your key together:
Claude or ChatGPT connected over MCP is a different thing entirely. When you connect Time Trakkr to your own Claude or ChatGPT, that assistant is already yours and already runs on your subscription — we never call Anthropic at all. Adding a key here has no effect on it. The two features are easy to confuse because both are "bring your own AI"; this page is about the assistant inside Time Trakkr.
It does not change what we store. Your conversations with the in-app assistant are still saved in Time Trakkr so you can scroll back through them, exactly as before. Using your own key changes whose Anthropic account the prompts pass through — not what Time Trakkr keeps.
It does not change Slack. The /track command runs fixed operations rather than asking a model anything, so it never used an Anthropic key in the first place.
Go to Settings → Assistant. You need to be an owner or an admin.
Paste your key and choose Save and verify. We make one small test call to Anthropic before storing anything — if the key is wrong, or the account has no credit, you find out immediately rather than discovering it when somebody asks the assistant a question next week.
Once saved, the key is encrypted and cannot be read back — not by you, not by us, not by the assistant. Settings shows the last four characters so you can tell which key is installed. If you lose your copy, generate a new one at Anthropic and save that instead.
You can point the assistant at a different Claude model. Leave it on the default unless you have a reason.
Natural-language entry and calendar drafts always use Haiku and ignore this setting. They run automatically for every member, every day, and a larger model there costs noticeably more without producing better matches.
Because the account is yours, most problems are fixed in your Anthropic console rather than in Time Trakkr. Time Trakkr will tell you which:
| What you see | What it means |
|---|---|
| "Your workspace's Anthropic key was rejected" | The key was revoked or mistyped. Save a new one. |
| "Your Anthropic account has no remaining credit" | Top up at Anthropic; the key itself is fine. |
| "Your Anthropic account is rate limited right now" | You have hit your own rate limit. It clears on its own. |
| "Anthropic is having trouble" | An outage at Anthropic. Nothing to fix at either end. |
We never quietly fall back to our key when yours stops working. You chose to run on your own account, and silently moving your prompts back through ours — and billing ourselves for it — is not a decision we should make for you. So when your key fails, the assistant says so and stops.
Time entry and calendar drafts behave differently, because they have a non-AI path: they fall back to Time Trakkr's own simpler matching. You still get a draft, it is just less accurate. Nothing is lost.
If a key is rejected outright, we stop trying and email every owner and admin — otherwise the only sign would be on a settings page nobody has open. Once you have fixed things at Anthropic's end, use Check again in Settings to clear the error without re-pasting the key.
Pause and use Time Trakkr's key keeps your key on file but sends AI usage back through our account — useful if you want to stop spending on Anthropic for a month without having to find your key again afterwards.
Remove key deletes it outright. That cannot be undone, since we hold no copy you can read.
The first message after switching is slower. Time Trakkr caches part of the assistant's prompt so repeat questions are faster and cheaper, and that cache belongs to whichever key made it. Switching keys starts a fresh one. It warms up again after a message or two.
One key per workspace. If you run separate production and sandbox workspaces, each has its own key and its own settings.
Updated 2026-08-23. Still stuck? Contact support or return to the Help Center.