First pass, one sitting

Sixty unread PDFs, and a review due in three weeks

Index the PDFs sitting in your downloads folder, ask the question you actually have, and read a structured map of cited claims. A typical workflow takes 90 to 180 seconds, so the afternoon goes on iterating rather than waiting.

Create your account

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

The agent building a mindmap of citation-backed claims from an indexed library — One instruction, a map of cited claims, in a minute or two.

The reason you are here

The hardest part is the blank page, not the reading

You have the papers. You have skimmed maybe a third of them. What you do not have is a structure: which claims belong together, which ones fight each other, and where your own question sits among them.

So the review keeps not starting. Reading one more paper feels like progress and postpones the part that is actually hard, which is deciding what the literature says.

A first pass you can argue with beats a blank page. It does not have to be right to be useful, it has to be structured and checkable.

How the first pass works

From a folder of PDFs to a structured map

Three steps. The first is the only one that takes real time, and it runs in the background.

Add the PDFs you already have
Upload them, or include items straight from Zotero with the plugin. Indexing runs OCR that preserves multicolumn text, tables, footnotes and captions, then writes structured English summaries of each passage.
Ask the question you actually have
Not a keyword. A research question, in your own words, with any domain-specific instructions the agent should respect.
Read a map, not a wall of text
You get branches of cited claims with confidence scores. Expand a branch, challenge a node, or ask for more evidence on the part that matters.

In the product

What an afternoon leaves you with

Indexed sources on one side, a map of cited claims on the other, and the conversation that produced it kept alongside.

Indexing papers into a knowledge base in Agent Bayes — Papers go into knowledge bases you attach to a project.

A developed mindmap of citation-backed claims after several agent runs — The map after a few passes, with citations on every claim.

What you can rely on

Why the first pass holds up

A pipeline, not a prompt

Your instruction moves through distinct stages: understand what you asked, retrieve and rank evidence, write cited nodes, then score whether the result satisfied the question and run another pass if it did not.

90 to 180 seconds per workflow

That is a typical run. Long enough to be doing real retrieval, short enough that you iterate rather than wait.

You set the depth

Ask for a quick answer, a full evidence workflow, or a deep sweep with more evaluation rounds. Pick the model and how much it deliberates. Speed, cost and completeness are your call each time.

Scope it with tags

Tag documents and restrict a run to a subset of your corpus, so a methods question only consults the methods papers.

It distinguishes argued from cited

Retrieval separates what a source argues itself from what it merely references, which is the difference between a citation that holds and one that embarrasses you.

Cancel and pay for what ran

Costs are estimated and reserved up front, metered in 0.25 credit increments, and only actual usage is charged, including mid-run cancellation.

What Agent Bayes does not do

It does not find the papers. You bring the library, which is exactly why it can only cite sources you have vetted.
It does not write the review. You get a structured, cited map to argue from, and the argument is the part that has your name on it.
It is not instant. Indexing takes real work per document, and a workflow takes 90 to 180 seconds. An afternoon is the honest unit, not a minute.

The first one is the trade at the centre of this product. No open-web search means no surprise citations, and it means curation stays your job.

Before you ask

Questions researchers ask first

How many papers do I need before this is useful?

Five is enough to see the shape of it, which is what the free plan covers. Thirty to sixty is where the map starts telling you things you had not noticed.

Will it find papers for me?

No. It is not a search engine, and that is deliberate. It reads the library you curate, which is why it can never cite something you have not vetted.

How long does indexing take?

It runs in the background per document, so you can keep working. Indexing an average 20 to 30 page PDF costs about 20 credits.

What if my papers are not in English?

Indexing writes its structured summaries in English whatever the source language, so a French, German or Japanese paper is searchable and citable alongside the rest.

Is a first pass actually any good?

It is a first pass, and it is honest about that. Every claim is cited so you can check it, thin evidence is reported as a knowledge-base limitation rather than padded, and the map is built to be edited rather than accepted.