Research Agent overview
The Research Agent (/agent) runs a bounded planning loop over existing tools (calculator, scenarios, simulation, causal, data inspection, retrieval, and related adapters). It is an orchestrator, not a replacement for professional judgment.
Research questions
State a concrete goal. The agent may propose tool calls; human-in-the-loop controls apply. Association-without-design can be an honest stopping point.
Sessions and persistence
- Authenticated sessions are owner-scoped in postgres mode.
- Browser restores a session id only after GET /api/auth/me verifies the user.
- Logout clears local conversation UI; PostgreSQL rows are not deleted by logout alone.
- Foreign session ids return NOT_FOUND.
Evidence and citations
Retrieved chunks are evidence candidates. Citations format evidence IDs — they do not invent authors, years, or DOIs. Attach agent artifacts to the notebook only when present; academic reports list evidence IDs without fabricating bibliographies.