Source: Gemini Deep Research agent docs

Google's guide to the Deep Research agent in the Interactions API: agent IDs, background execution, streaming, collaborative planning and tools.

Contents
  • What: official API documentation
  • Author / publisher: Google
  • Date: read 2026-11-14 (living page)
  • Original: aistudio.google.com/docs/deep-research (formerly ai.google.dev/gemini-api/docs/deep-research)

Summary

Gemini Deep Research is exposed as a managed agent, not a model: you create an interaction, it plans, searches, reads and iterates for minutes, and returns a cited report.

Key claims

  • Agent IDs (as of 2026-11): deep-research-preview-04-2026 (tuned for speed and streaming to a UI) and deep-research-max-preview-04-2026 (tuned for maximum comprehensiveness). Preview status.
  • Only available through the Interactions API (client.interactions.create(agent=..., input=...)), not generate_content; follow-ups use previous_interaction_id.
  • background=True is required; poll interactions.get(id) until completed or failed.
  • Streaming: stream=True plus thinking_summaries: "auto" yields thought, text and image deltas; streams can drop (600-second timeout) and resume with last_event_id.
  • Default tools: google_search, url_context, code_execution; optional file_search over your document stores and mcp_server (with headers and allowed_tools).
  • collaborative_planning: true returns a proposed plan for the user to refine before research starts.
  • Optional charts (visualization: "auto"), image and PDF inputs, and report structure steered by the prompt.

Feeds

Deep research agents

Suggest changes →