Closed beta · Windows

AI starts every session from zero. Borje starts it where you left off.

Persistent project memory for AI CLI tools: human-readable, git-friendly markdown. On your disk, with every tool.

Memory is not a database. It's a file you can read.

Everything Borje keeps is plain markdown: open it, read it, fix it, commit it. You don't guess what the AI knows about you. You see it.

.borje/active.mdindexed locally

# Active

## Now: payment flow refactor
- Added Stripe webhook idempotency key
- Decision: retry queue on Postgres, not Redis
  (one less dependency, [[decisions]] 2026-07-02)
- Next: webhook signature tests

→ This context loads when Claude Code opens.

How it works

Tracks your work

Your decisions, tasks and session traces accumulate as plain markdown under .borje/. Open, read, edit any time.

Indexes on your device

Embedding, search and context building run locally. Your memory never reaches Borje's servers; only account verification does.

Starts every session warm

Claude Code, Codex, Antigravity, Copilot. Whichever you open, your project context comes preloaded. No re-explaining from scratch.

Same question. Two sessions.

A session without memory reintroduces itself every morning. A session with Borje already knows why yesterday's decision was made.

without borje

$ claude
> Why did we move the retry queue to Postgres?
● Scanning the codebase… (12 min)
I can't find the reasoning in the code.
Can you explain why Redis was dropped?
— you're re-explaining the context again

with borje

$ claude
◆ borje: context loaded — payment flow refactor
> Why did we move the retry queue to Postgres?
One less dependency (decision, Jul 2).
Next up is webhook signature tests. Starting.

Where does your memory go? Nowhere.

Indexing, embedding and context building run on your device. The only thing sent to the server is account verification. Your files live on your disk, in your git, in your backups: wherever you put them.

Memory: Plain markdown, versioned in git, on your disk
Processing: Local embeddings, no LLM calls in the loop
Tools: Claude Code · Codex · Antigravity · Copilot
Platform: Windows (closed beta), desktop app

Frequently asked questions

Does my memory go to the cloud?

No. Indexing, embedding and context building run on your device. The only thing sent to the server is account verification. Memory files stay on your disk and in your git.

Which tools does it work with?

Claude Code, Codex, Antigravity and Copilot. When you switch tools, Borje carries your memory with you: they all read the same .borje/ folder.

Why markdown and not a database?

Because you should be able to read your own memory. Markdown opens, edits, commits and backs up. Even if Borje shuts down, your files stay with you; no locked format.

Is the beta paid?

No, the closed beta is free. Applications are approved one by one; if you actively work with AI CLIs, you get priority.

Can I use it outside Windows?

Not yet. The beta is Windows-only; macOS is on the roadmap. The memory format is portable, so migration will be painless.

The closed beta moves in limited batches.

Applications are approved by hand, one by one. If you really work with AI CLIs, you move up the list.

Apply for beta