AI triage

AI triage reads an incoming issue and proposes the boring-but-important metadata: a priority, the right team, sensible labels, and likely duplicates. It is available on every plan (the Free tier includes a monthly AI-call allowance); on Business and above you can run it through a bring-your-own gateway key so the spend and the model choice are yours.

Human in the loop

Proposals are suggestions, not silent edits. A person accepts, tweaks, or rejects them. Anything an agent does that is destructive (bulk changes, deletes) is held behind the same propose-then-approve guardrail as automations and recorded in the audit trail — there is no path for the model to mutate your workspace without a check.

Budgets and isolation

AI usage is gated by a per-workspace budget: a request is pre-checked against the budget before it runs and the spend is recorded after. Like everything else, AI runs inside your tenant boundary — the model sees only data the requesting member could already see.

What it is not (yet)

AI triage is assistive, not autonomous. It does not close issues, ship code, or make planning decisions on its own. Treat its output as a fast first draft a human signs off on. Deeper autonomous agent workflows are an area we are actively building — see the Agents page on the marketing site for the current state.