Research Agents

Automatically gather information from multiple sources and write the findings into your document.

Overview

Research agents handle the work of collecting information from different places and synthesizing it into something useful. Instead of searching your codebase, checking your team's guidelines, reading a linked article, and then writing it all up yourself, you describe what you want researched and the agent does all of that in one go. The results land directly in your brief, blueprint, or method — organized and ready to work with.

Research agents are part of the AI assistant and activate automatically when you ask for research. You don't need to choose anything from a menu — the assistant handles the right approach based on what you're asking for.

How It Works

  1. Trigger a research request — In the chat panel, describe what you want the agent to research. For example: "Research how we handle authentication in the codebase and update the brief" or "Find our approach to this feature in the methods library and add a summary."

  2. The agent runs parallel searches — Depending on what you asked, the agent may search your team's blueprint library for organizational context, your methods library for process patterns, your connected codebase, any URLs you've mentioned, files you've attached, and linked project management issues. These run in parallel to keep things fast.

  3. Findings are synthesized — Once all searches complete, a synthesis step combines the results and writes them into your document in the appropriate section, or returns a summary in the chat if you asked a question rather than requesting a document update.

  4. Steps are visible in the chat — You see each step labelled as it runs: "Searching organizational context," "Researching codebase," "Extracting URL content." When everything completes, the final result appears and the steps collapse automatically.

Research Flows

Different research scenarios trigger different flows:

Full research activates when you ask for a comprehensive investigation. It searches your blueprints, methods, codebase, any provided URL, attached files, and linked project issues — all in parallel — and then synthesizes everything into your document. This is the most thorough option.

Quick research answers a focused question without modifying your document. Use it when you want to know where something is implemented or how a specific pattern works, without committing the findings to the brief.

URL-to-document extracts content from a specific webpage and adds it to your brief. Paste or mention a URL in your message, ask the agent to add it to your brief, and it fetches the page, pulls out the key points, and writes them in.

Meeting notes processes a meeting transcript or notes file and extracts decisions, action items, and discussion points, then structures them as a meeting summary section in your brief.

Linear research pulls in context from linked project management issues when you reference an issue ID in your message. The findings are incorporated alongside any other research sources.

Key Capabilities

  • Parallel execution: Multiple searches run at the same time. A full research request that covers blueprints, methods, codebase, and a URL runs them concurrently rather than one after another.

  • Organizational context awareness: The agent checks your team's blueprint library for relevant company guidelines, standards, and principles before writing. If your blueprints say how your team approaches a particular type of work, the agent incorporates that framing.

  • Process pattern matching: The agent searches your methods library for relevant workflows and procedures. If your team has documented an SOP or implementation pattern that applies, it gets included.

  • Codebase research: When your workspace has a codebase connected, the agent can search it for implementation patterns, relevant files, and architecture decisions.

  • Source attribution: URL content added to a brief includes a reference to the source page.

  • Focused or comprehensive: Ask for a quick lookup or a deep dive. The agent matches its scope to your request.

Tips

  • The more specific your research request, the more focused the output. "Research how we currently handle rate limiting in the API layer, following our backend guidelines" gives better results than "research rate limiting."
  • Mention a URL in your message if you want the agent to pull from an external source. The agent detects URLs automatically and includes URL extraction in the flow.
  • Attach files to your brief's context before asking for research. Meeting transcripts, design specs, and reference documents become source material the agent can incorporate.
  • Use quick research when you want to check something without adding it to your document. It's faster and keeps your brief clean.
  • If you've already done a research run and want to update the brief with new information, ask the agent to "add to" or "update" a specific section rather than running a full research pass again.

Related