The AI brain¶
The brain is optional. Every screen works without it — it only adds three conveniences, and each is something you trigger, never a background surprise.
What it does¶
- Next action — for a task, a one- or two-sentence concrete next step, precomputed when the task's items change so it's there when you open it. Backfill them all from Admin → AI.
- Item → task matching — on demand (Match on an item) or in bulk (Match all in catch-up), it suggests which existing tasks an item belongs to. It only proposes above a confidence floor — no match is a normal, common answer.
- Bucket suggestion — for an Uncategorized task, a bucket (existing or new) with its reasoning. It only suggests; enrichment will adopt a suggested bucket automatically only when that bucket already exists and it's confident — it never invents one for you.
Which backend it uses¶
HQ picks the first credential it finds, in this order — so a machine you've already logged into needs nothing stored:
- an API key you saved in Admin → AI (kept in the OS keychain);
- the environment (
ANTHROPIC_API_KEY/ANTHROPIC_AUTH_TOKEN); - a logged-in
ant auth loginprofile; - the local
claudeCLI on your PATH (subscription pricing — HQ shells out to it); - none — the AI features are simply unavailable, and everything else keeps working.
The Admin → AI panel shows which of these is live and the model in use.
What it's sent¶
Only your own connected-tool content — task titles, descriptions, buckets, tags, and each item's source, label, context and extracted ticket references. Never credentials. Item content is treated as context to reason over, not as instructions.