Adds the chat icon to the app header so members of this org can use the AI Assistant.

Which model this org's AI uses (chat + policy review). Haiku routes to a separate Foundry deployment, so it has its own tokens-per-minute quota and lower cost. Usage is billed at the chosen model's rate.

Hard cap on Claude spend per calendar month. Assistant pauses $1 before this cap to prevent overrun. Resets at 00:05 UTC on the 1st of each month.

Maximum characters a user can send per chat message. Default 1,500. Lower values reduce cost and abuse.

Stops a single question from pulling in a runaway amount of data (e.g. a whole month of archives). Uncheck to turn the guards off entirely.

Ceiling for one question (model context window is 1,000,000). Default 300,000 — high enough that normal questions never hit it, low enough to stop an accidental data dump.

Largest single data result the assistant will take in (anything bigger is trimmed). Default 60,000.

When on, the assistant can only read and act on sections in the tab the user is currently viewing. It can still navigate the user to other tabs/sections, but cannot touch their data until they go there. Also trims what the AI loads per question (lower token use).

Turns on AI policy review (flags gaps against regulations & standard of care) for this org's Policies & Procedures. Billed separately — its spend does NOT count toward the budget above.

Turns on AI normalization of fetched job postings (specialty, shift, pay extraction) for this org's Job Board. Uses a dedicated model deployment (FOUNDRY_JOBS_DEPLOYMENT). Billed separately — its spend does NOT count toward the budget above or the Policy AI budget.