For agent developers
Every agent picks up
where the last one left off.
Claude Code writes it, Codex reads it, Cursor continues it — one memory over MCP, with provenance on every write so two agents stop undoing each other's work. Not a CLAUDE.md you hand-sync, not a vector index you babysit: verbatim storage, full version history, and an audit log you can actually read.
One command, real OAuth
claude mcp add --transport http skeleton … then a browser login. No config files, no pasted tokens — any MCP client pairs the same way.
Provenance on every write
Every memory records which agent, which session, which timestamp. When Codex and Claude disagree you get a proposal to resolve — never a silent overwrite the next agent acts on.
Yours to audit, yours to leave
Verbatim storage, hash-chained version history, and a one-command export to JSON or markdown. Easy to leave is what makes it safe to build on.
Not a promise — a ledger.
- Every read and write is hash-chained and timestamped. Open the log; it can't be quietly rewritten.
- Export everything as JSON or markdown in one command. No lock-in, no export ticket.
- Verbatim storage — what you said, kept as you said it. Inspect the raw record any time.
Wire it into your stack in a minute.
Free to try. Poke the audit log and the export first — that's the point.