Activate AI only after a policy review

The KB assistant remains disabled unless separately approved; new framework quota/circuit adapters have not been installed here.

Steps

Read the application AI runbook first. Configure one environment's API key, supported model, approved data-sharing policy, conservative price ceilings, and daily budgets. Then explicitly enable the provider and opt selected developers into ai.ask.

current account permission → explicit sharing consent → visible document retrieval
                          → cost/quota reservation + safe audit
                          → one bounded provider request, no tools
                          → closed schema + current citation verification

Do not paste credentials, customer data, or production logs into a question. Only approved visible excerpts enter context; no arbitrary paths, URL fetches, shell, SQL, writes, or deployment tools exist. Optional general suggestions are explicitly labeled unverified rather than framework facts.

Verify

Fake-provider checks precede separately approved live evaluation. Valid citations establish excerpt provenance, not a mathematical proof that every model sentence is correct. Human review remains necessary for coding/security decisions. See AI mechanics.

Editorial provenance

Source: resources/guides/ai-policy.md. Source SHA-256: 7e9e337e6564e7f790789ab53ebb12b5885fd61c5db7874f2a9d60727d09aadc.

Knowledge-base application source; editorial checkpoint 2026-10-20. See candidate status and adoption boundary.