Stop re-explaining yourself. Build your context once.
You've explained your project to ChatGPT a hundred times. Mara interviews you once and writes the file your AI reads before every reply — so every new chat already knows your company, your projects, your voice.
Starts as a handful of markdown files. Grows as you drop in articles, transcripts, and notes. Read by any agent that can read files.
Run the interview from inside your agent.
Mara is a plain HTTP protocol — your agent does the interview, then writes your context straight into your filesystem. Paste one prompt into the Claude, Codex, or Cursor app and you're done.
Drop CLAUDE.md next to your folder. Every new chat reads it first.
AGENTS.md lives at your project root. Codex picks it up automatically.
Same context as a Cursor rule. .cursorrules → context.md → projects/.
cd ~ && claude
> Interview me with Mara's protocol from
mara.run/api/public/interview-spec, then POST the
transcript to /api/public/bootstrap and write the
files into ./context/cd ~ && codex
> Fetch mara.run/api/public/interview-spec, run the
interview, then POST to /api/public/bootstrap and
drop the resulting files into ./context/# In Cursor's chat (cmd-L):
@web mara.run/quickstart — follow the protocol,
interview me, then write your context files into ./context/
in this workspace.Every AI you already use can read it.
Your context is plain markdown. No plugin, no API, no vendor. Paste context.md into ChatGPT. Drop CLAUDE.md next to your repo. The file format is the integration.
Point it at the folder. It reads CLAUDE.md, then context.md, then projects/. Your codebase finally has a manual.
Paste context.md into a new chat (or save it as custom instructions). Every thread starts with you already understood.
Open your context as a workspace, or drop CLAUDE.md / AGENTS.md / .cursorrules — same file, same context, no porting.
Plain markdown means the next agent that ships in 2027 reads it too. No SDK. No vendor lock. The file is the integration.
Every page is a .md file on your computer.
No cloud. No database. No subscription. Obsidian opens the folder; so does Finder, VS Code, ChatGPT, and Claude Code. If any of them disappeared tomorrow your context would still open in TextEdit. That's the point.
Your AI's onboarding doc, finally written down.
Articles you want it to read. Meeting transcripts. Business context. Notes. Decisions you made last week. All in plain text. Paste context.md into any new chat and stop re-onboarding it from scratch.
Same spine for everyone. The branches Mara writes depend on you.
Same shared spine for everyone. The yellow lines are the add-ons Mara branches into based on what you tell her in the interview.
You're the context, and you're tired of typing it.
Every new ChatGPT thread: a paragraph about your company, your team, this quarter, what you're trying to do, the constraint that matters. Six months of re-pasting the same context because there's nowhere it lives. Your AI is smart. It just doesn't know you.
Mara's move · Mara names this back to you in turn one. Your context starts with context.md — the one file you paste once and stop re-explaining.
Stale context is worse than no context.
A doc you wrote in January is wrong by April. Your AI confidently uses outdated info, and you spend the chat correcting it. Your context only earns trust if Friday-you keeps it current — which takes 20 minutes if it's small and zero minutes if it's an app you stop opening.
Mara's move · Every starter ships with weekly-review.md whose third prompt is 'what changed about my context this week?' That's the heartbeat.
People start with structure instead of a context file.
Day one: build PARA folders. Day two: nothing to put in them. Day fourteen: closed. Your context dies before it gets paste-able. Structure is for maintenance, not for starting.
Mara's move · Mara skips PARA on day one. A handful of files, one paste-able context block, one seed project. Structure is the upgrade path, not the entry point.
Three people, one wound: my AI doesn't know me yet.
Every ChatGPT thread already knows your company.
Stop re-pasting your team, your quarter, and your stakeholder map. context.md does it once. weekly-review keeps it current.
See your starterYour AI knows your draft, your sources, your voice.
Twelve voice memos, three drafts, one through-line your AI keeps forgetting. context.md and projects/brief.md fix that.
See your starterPlain markdown your agent reads — Claude, Cursor, Codex.
CLAUDE.md, AGENTS.md, .cursorrules sit next to your repo. Every new session starts with the codebase already in its head.
See your starterOne-time. The interview, your context, your folder. Yours forever.
No subscription. No seats. No "premium" tier locked behind a paywall. We charge a dollar because charging zero attracts agents abusing the endpoint — and because your context is worth more than a dollar.
Not a notes app. Not a sync service. Not a course. Not a forever-chat AI brain.
We hand you a folder of markdown and walk away. The system is yours. The retrieval is yours. The Friday ritual is yours. Mara is the person who got you off the blank page — not the place your brain lives.
We drop the conversation the moment your starter is generated.
We keep a hash and a timestamp. That's it. No transcript retention by default — your captures, your wound, your manager's name, gone. "Let Mara remember you" is opt-in, and it's not built yet. When it is, you'll see a checkbox, not a default.