Introduction

Agent Bayes is an AI research assistant built around a shared, interactive mindmap. You bring a library of papers. The agent helps you explore, synthesize, and structure what's in them, and you direct every step.

What you can do

  • Instruct the agent in natural language (e.g., "expand the Bronze Age collapse section with theories about Sea Peoples"). It researches your sources and writes citation-backed nodes into the mindmap.
  • Search your library semantically and attach citations to your own nodes without any agent involvement.
  • Iterate by continuing the conversation: refine claims, restructure the mindmap, reorganize into chapters, or polish prose.
  • Trust what you see. Every substantive claim is backed by a citation from your knowledge base that can be verified. The system never fabricates sources, and contradicting viewpoints are preserved as sibling nodes rather than averaged away.

How it works at a glance

  1. You create a knowledge base and add papers to it. Upload PDFs directly, or sync via our Zotero plugin.
  2. You create a project and attach one or more knowledge bases to it. A project contains many mindmaps, and each mindmap can draw on any of the project's knowledge bases.
  3. You create a mindmap and shape it together with the agent. Write your own nodes, or instruct the agent to research and add nodes for you.
  4. You write instructions in natural language. A multi-agent system routes the request, retrieves from your knowledge bases, synthesizes citation-backed claims, and applies edits to the mindmap. The mindmap is the single source of truth for your research.
  5. Typical workflows complete in 90–180 seconds, streaming updates to your mindmap as they happen.
  6. After each round of research and editing, the system analyzes gaps and derives follow-up tasks. This loop continues until your instructions are satisfied, or no further evidence can be found.

Who it's for

Master's and PhD students, postdocs, researchers, analysts, and writers who work with bodies of literature and want a structured, citation-grounded workspace, not a chat window that forgets where the claims came from.

What's next