Map and synthesize arguments from your PDF library

Synthesize clear arguments, organize your research visually, and keep every citation traceable, all in one persistent workspace that holds it together.

Start now

Works with your PDFs and your Zotero library

Build a better argument, and stand behind every line of it

Agent Bayes helps you find, organize, and synthesize evidence across your own library. You stay in control of how the evidence is interpreted and used.

Build and refine the argument

Draft claims that pull from several sources
Grow the map branch by branch, or restructure it yourself
See where the literature goes quiet, and where it conflicts
Refine over weeks with snapshots and diffs

Write it clearly

Turn any branch into a cited prose discussion
Use the map as the outline for your own draft
Tighten wording that overstates what the sources say
Read foreign-language passages in English

Get the citations right

Open the exact passage and page behind any claim
Printed page numbers, ready for your bibliography
Score how well your citations support your wording
Every retrieval stays inside your own library

Start researching for free

Free to start, no card. 200 credits when you verify your email.

Every claim shows its source

Click a claim and the passage behind it opens at the page it came from. You read the wording in context, in the original PDF, before it reaches your draft.

Citations carry the printed page number, ready for your bibliography. Retrieval stays inside the knowledge bases you attached, so a claim never arrives from a source you did not choose.

The map is yours to edit

Every node the agent writes is ordinary text. Rewrite it, move it with the keyboard, or delete it. Add your own claims anywhere in the tree and pin citations to them from search.

Nothing is locked. Undo and redo cover every change, yours and the agent's, so you can restructure the argument freely without losing work.

Real maps, published

See what a finished map looks like

Three maps built in Agent Bayes and published as public links. Open one, expand a branch, and click a citation to read the passage the claim came from.

AI Agent Safety & Alignment

The 2026 literature on agent safety, from misalignment risks through alignment methods to evaluation. One node holds a deliberate error, left in to demo AI verification.

287 nodes26 sources270 citations

Open the map

Reward Hacking

How agents maximize a proxy reward that drifts from the intended objective, with the formal treatments, mitigations, and detection work that followed.

60 nodes8 sources67 citations

Open the map