Usage & troubleshooting
Find a request, understand a limit, and choose what to fix before retrying.
Find out what happened
Open Logs for workspace activity. To focus on one agent identity, open Profiles, select its name and choose Activity.
Check the action, time and outcome against what the tool attempted. A Profile can have several keys; each event shows the label of the key that made the call where one is recorded. Human changes can also appear in workspace activity, so not every event belongs to an agent key.
If a tool reports an error, keep its status code and message for diagnosis. Never include its secret key in a support message, screenshot or shared log.
Read daily usage and limits
Open Settings → Usage to inspect usage meters. Profile creation includes Daily budget (tokens) and Asks limit (per day). All keys using a Profile share those settings; adding a key does not add another allowance. Open a profile to change either number later: lowering one below what the profile already used today blocks new calls until you raise it or the counters reset, and the change appears in the activity log.
A token is about 4 characters of English text, roughly three-quarters of a word. Three rough examples:
- A short question.
What is our refund policy?is 26 characters, so 7 tokens, plus whatever evidence comes back. - A typical ask. Question plus a few evidence passages usually lands in the low hundreds of tokens.
- An upload. A five-page document of about 15,000 characters costs about 3,750 tokens when it is indexed.
For scale, Ember's 300,000 daily tokens are about 1,200,000 characters, roughly eighty five-page documents indexed every day. That is far more than a few hundred asks will spend.
So for ordinary questions and answers, you will usually hit the ask count first. A few hundred daily asks at a few hundred tokens each stays far below the workspace token allowance. Token budgets matter instead for heavy evidence, and for uploads and imports, whose converted text counts against the workspace allowance when it is indexed.
Daily counters reset on the UTC day, which may differ from your local midnight. A token budget and an ask count are different limits: a request can hit one before the other.
A 429 response can be a short-term rate limit, a Profile budget or a workspace allowance. Read the message and the meters before retrying. Immediate repeated requests do not replenish a daily allowance.
Diagnose a failed request
| Symptom | What to check |
|---|---|
| Cannot reach the API | Check the network with takibi version. The version check needs no key, so test a search afterward to check authentication. |
401 | Check that the tool can read its complete key, and whether the key was rotated or revoked or the Profile paused. Also check whether the requested action needs a human session. |
403 or 404 | Check the collection ID, document grants and required capability. Some out-of-scope resources appear as not found. Do not broaden access unless the tool should actually have it. |
409 | Refresh the task before retrying. Another Profile may have claimed it, or its state may have changed. |
429 | Read whether this is a rate or daily-budget limit. Slow down rate-limited requests; daily budgets reset on the next UTC day. |
| Server failure | Keep the error message, check service availability and retry after the underlying issue is resolved. Check whether a previous change completed before repeating it. |
The CLI exits with 0 for success, 1 for transport or API failures, and 2 for usage errors. For usage errors, run takibi --help and check the command arguments.
A successful request returned no evidence
A response with no evidence is not an authentication or network error. The JSON marks it abstained: true. Check that the source is indexed, the Profile can read its location, and the question is covered by that source.
Try a specific search phrase and inspect the converted text. If the knowledge is missing or unclear, update the source instead of repeatedly asking the same question. See evidence fields and missing-topic suggestions.
Included AI usage
Plans include AI usage for companion runs, chat with Emi and Takibi web search. The AI usage meter in the sidebar and Settings → Usage show how much is used and when it resets.
Included AI usage resets with your workspace's billing period, on the date shown there, not on the first of the month. A trial includes a smaller allowance; the full one starts with your paid plan. A banner warns you when 80% is used.
When it is used up, chat pauses until the period renews, and new companion runs pause too, unless you turn on starting new runs on your own OpenRouter key. A run on a model that bills your key does not use the included allowance for that model; Takibi web search in the run still counts. See models and who pays and what chat costs.