MIT licensed · Run it yourself · Your key

Every AI knows the same things. This one knows what you read.

One brain per subject. Each holds a position, the evidence under it, and a date on every claim.

This instance holds one person's brains, so the app asks for their account. Reading them needs nothing: point any AI client at it.

Three things you do

Drop

Feed it a source

Paste a link and it opens the page. One card comes back: where it lands, what is new, what repeats, and every claim that contradicts what you hold. You rule each one, then it writes.

Ask

Ask in plain words

The first sentence answers. Numbers sit inside the answer, sources on one line underneath. A follow-up question reads against the ones before it.

Connect

Use it from your own AI

One address makes any MCP client a reader. A second one lets it feed. Your client does the thinking, so a drop through it spends nothing here. How to add it.

Run your own

The whole thing is MIT licensed and public. Two ways to run it, and a brain exports to plain markdown either way.

As a web app

A Convex deployment holds the data. Any static host serves the pages.

git clone https://github.com/jlasne/brain
cd brain && npm install

# a deployment of your own
npx convex dev

# the key that pays for reading and rewriting
npx convex env set OPENROUTER_API_KEY sk-or-... --prod
npx convex env set KEY_SECRET "$(openssl rand -base64 32)" --prod
npx convex deploy

# optional: one owner, no sign-ups
npx convex env set ONLY_ACCOUNT yourname --prod

# optional: video links fetch their own transcript
npx convex env set SUPADATA_API_KEY sd_... --prod

Then serve app/ as the site root. The repo carries a vercel.json that already does it. Open the app, take a name and a password, and create your first brain.

As files on your machine

No server, no database. A brain is a markdown folder you can open in any editor.

cp -r brain/skill/brain ~/.claude/skills/brain

Then ask Claude Code for your first brain. Same three actions, same rules, same export format as the web app.

What is in the repo

Path	What it holds
`PROTOCOL.md`	Every rule the brain follows, numbered: how a source is filed, how evidence compresses, what the receipt has to prove.
`convex/`	The server. The routes, the four drop steps, the MCP server, the schema.
`app/`	The pages. Each one is a single HTML file with no build step.
`skill/brain/`	The Claude Code skill, for running the whole thing as files.
`templates/`	The markdown a brain exports to: the map, a concept, a note.
`scripts/`	The checks. One reads every page, one drives a whole drop with no network.

Nothing here locks you in. A brain exports to markdown, and the protocol that produced it sits in the same repo, so it can be rebuilt anywhere.