Not stored: looks like an API keySecret-shaped text is rejected before it reaches the database.
The idea
What one agent learns, every agent knows.
A chat ends, you switch tools, a teammate picks up the work — and the context is gone. With Brainkeep your agents write down what matters as they go and read it back before they start, in any client and for your whole team.
Your agents
Claude Code
claude.ai
Codex
Antigravity
Any MCP client
BrainkeepProjects, revisions and authors in one memory you control
Your team
AAnaOwner
LLeoRead and write
MMaraRead only
How it works
Save, revise, recall.
1remember
Save
Agents record decisions, discoveries and progress on their own, guided by a short protocol the server sends — no long prompts.
Before work, any agent pulls compact context for the project — in any client, for anyone you share it with.
context(
project: "checkout-api",
compact: true
)
Features
A small server with a serious memory.
Written in Rust
Native Rust, end to end.
One binary with no runtime and no garbage collector, embedded SQLite and the web panel built in. Nothing else to install: no Node, no Python, no database server. It runs even on a Raspberry Pi.
Language
Rust, no GC
Storage
Embedded SQLite
Panel
Built into the binary
Runs on
macOS · Linux · Raspberry Pi
Revisions and authorship
Every memory keeps its full history, with verified authorship on each revision.
Built for teams
Invite people by email, share projects read-only or read-write, revoke access instantly. Google sign-in over OAuth.
Small context footprint
Compact responses and a short memory protocol, so agents save on their own without bloating the prompt.
Safe by default
Passwords, keys and tokens are rejected before storage. Private files and per-user request limits.
A panel for people
Search, filters, history, diffs and live updates. Edit with conflict handling, archive and restore.
Backups and migrations
Consistent backups on a systemd timer and schema migrations with snapshots, so upgrades are safe.
Self-host
Your memory lives on your machine.
Run it locally for yourself in a minute, or on any Linux host with systemd behind HTTPS for your whole team — a Raspberry Pi is enough.