Deep Research agent
Run a multi-step web research job: plan → search → cited draft → staging inbox → confirm import. Works in the background while chat stays usable.
Before you start
- Signed in on Hub with a Mind notebook.
- Ops enables MIND_RESEARCH_AGENT=1 on the API (soft-first; default OFF — 404 until enabled).
- Pro or trial with remaining monthly research quota (shared with manual Deep Research).
Steps
- 1
Open Mind → notebook → Deep Research panel (Hub) or use Deep Research in Sources / Add Sources.
- 2
Enter a goal → Run agent (or Build plan first, then Run agent). A fixed progress dock appears; chat and Studio stay usable.
- 3
Wait for staging: review Cited vs Uncited candidates, toggle checkboxes (take/discard), then Confirm import.
- 4
Confirm import ingests selected URLs as sources and creates a research_report (or briefing fallback) in Studio with research_run_id lineage.
- 5
From chat: Find sources escalates to the agent when the flag is on; coverage-gap banner offers Run Deep Research for thin notebooks.
- 6
Cancel from the dock if you change your mind; quota was already counted on start (same meter as manual Deep Research).
Common mistakes
- This is NotebookLM-style Deep Research — not an arbitrary cloud computer or Python sandbox.
- Preview staging before import; discard low-quality SEO hits.
- 402 means monthly research quota exhausted — upgrade or wait for the next cycle.
- 404 research_agent_disabled means the API flag is still off; manual plan/search in Deep Research still works.
- Imported sources and reports show research_run_id in preview for traceability.