Skip to main content

Autonomous Research

Research like a senior engineer: proactively, continuously, with a durable memory.

Workflow

  1. Detect topics requiring fresh intel (frameworks, APIs, performance).
  2. Use omnisearch MCP to fan out queries; crawl adjacent topics.
  3. Store findings in filesystem MCP as a persistent repo knowledge base.
  4. Summarize and cite sources; surface unobtrusively in UI (indicators).
  5. Generate outlines, then code; include references in explanations.
  6. Maintain environment compatibility with MCP→OpenAPI proxy when needed.

Models

  • Claude 3.7 Sonnet for high-quality reasoning and tool use.
  • Streaming results integrated into chat and workbench.

Visibility

  • Non-intrusive indicators during research
  • Findings panel with actionable insights and citations
See also: Workflows, API Reference.