Blog
From the Agent Bayes team
Product updates, research workflows, and how we think about building tools you can actually trust with your literature.
Connect Claude Code to your indexed papers, figures included
The Agent Bayes MCP server lets Claude Code or any MCP client search your papers, get cited answers, and build mindmaps. It can now also look at PDF pages and the images on mindmap nodes, so maps and charts are no longer blind spots.
Guy Zana
September 25, 2026 · 7 min read
Gemini 3.8 Flash, GPT-5.4, or GPT-5.6 Terra for cited research synthesis?
We benchmarked Gemini 3.8 Flash, GPT-5.4, and GPT-5.6 Terra on citation-grounded synthesis over 150 indexed archaeology papers. Gemini weaves twice as many sources into each paragraph as Terra, and Terra's sentences stay closest to what the cited passage states. GPT-5.4 sits between them at two to three times the cost.
Guy Zana
September 8, 2026 · 12 min read
Most AI research tools are not competing with each other
We compared thirteen of them. The useful split is not which one is smartest. It is what corpus each tool reads, what its citations point at, and what you are left holding when you close the tab.
Guy Zana
August 3, 2026 · 9 min read
How to build a passage-level evidence matrix for a literature review
A step-by-step guide to coding passages, preserving source context, comparing themes across papers, and turning extracted evidence into claims you can trace back to the page.
Guy Zana
August 2, 2026 · 14 min read
We gave our research agent a glossary it builds from your papers, not the dictionary
Domain terms are local and overloaded. A small terminology graph gives a research agent an inspectable, corpus-relative glossary: top supporting passages and neighboring concepts, as evidence to reason over, not a canned definition to paste.
Guy Zana
August 1, 2026 · 10 min read
Our Customers Now Get 3x More Indexed Papers, for the Same Price
We moved the pipeline that reads and indexes your PDFs to OpenAI's gpt-5.6-luna. Indexing costs about 60% less, runs in under half the time, and we measured it against our own evaluation loop before shipping to make sure nothing got worse.
Guy Zana
July 31, 2026 · 2 min read
GraphRAG uses five LLM indexing stages. We use one extraction pass.
One LLM pass extracts scholarly terms and scores how informative each passage is. LLM-reconstructed semantic chunks make co-occurrences less noisy, while agents resolve meaning from evidence at query time and no glossary definition is committed.
Guy Zana
July 15, 2026 · 13 min read
Imposter syndrome in academia: what the research says, and where our AI misread it
A citation-backed tour of what the research actually says about imposter syndrome in academia: how common it is, whether it is even a real condition, and the one place our AI research assistant got a citation wrong while mapping the literature.
Guy Zana
July 5, 2026 · 7 min read
AI should help researchers think deeper, not think less
The useful role for AI in research is not replacing judgment. It is helping a human stay immersed in a problem, keep evidence and disagreement visible, and build knowledge that compounds over time.
Guy Zana
July 2, 2026 · 11 min read
Token accounting and cancellation in durable agent workflows
Closing an LLM stream is not the same as stopping the work behind it. Durable agents need cancellation paths that settle application state and credit accounting before the UI declares victory.
Guy Zana
July 2, 2026 · 9 min read
Introducing Agent Bayes: research you can actually trust
Why we built a multi-agent research assistant around a shared, citation-backed mindmap, and how it fixes what generic chatbots get wrong about working with literature.
Guy Zana
July 1, 2026 · 8 min read