Help & agent surfaces
How humans and automated clients use competitor-intel — same HTTP APIs, same auth.
In-app AI (Market History §E)
Generate insights posts chart JSON plus your filter tuple to /api/market-history/insights. Section E opens with quick-loop chips (ten angles) that each set insightHint; use plain Generate for no hint. The system prompt is editable in the repo at docs/prompts/market-history-insights-system.md. Column semantics are loaded from metadata_column_descriptions in Neon when available. Open Market History to run it.
External SQL Q&A agent
Read-only Postgres role ai_readonly — full playbook in the repo file docs/AI_QA_AGENT.md. Same database as the web app; no writes from that role.
HTTP API catalog (script parity)
Every route an operator can hit from the UI is listed for agents and scripts in docs/AGENT_API.md. Use the same base URL and Basic auth as a browser session when Vercel env vars require it.
RMS pricing workflow
Use Command /pricing for the summary queue, then open RevMan for the actual pricing decision. RevMan shows recommended ask, target signed rent, DOM checkpoint, drop schedule, concession stance, M20 floor handling, and regulatory framing. Compliance, Yardi, Funnel, Forward Book, and Move-Outs are supporting evidence surfaces — not competing pricing engines.
Agent-native audit
Fit-for-purpose scorecard (how we measure this repo): docs/AGENT_NATIVE_SCORECARD.md.