every ai shares one project memory
claude, codex, cursor — any mcp client reads and writes the same living project memory: goal, active work, blockers, decisions, open questions. pick up exactly where any of them left off.
alpha · local-first · mcp-native
stop re-explaining your project to every ai.
your AIs share one project memory, in files you own — and you can always see, veto, and undo what they know.
npx @aihq/ai-continuum-setup
01 · why
one memory. every ai. your rules.
visible, vetoable, undoable
every write lands with an immutable receipt. history, authority labels, and one-command undo on everything your ai knows. no black box, no silent edits.
local-first, hosted optional
your memory lives in plain files, sqlite, and git on your machine — with zero default telemetry. hosted convenience is optional, never required.
02 · how it works
three steps, then forget it.
-
step 01
install
npx @aihq/ai-continuum-setupinstalls the mcp server and cli. -
step 02
init your vault
the installer creates a local memory vault for your projects — plain files you own, in a directory you choose.
-
step 03
connect any mcp client
point claude, codex, cursor, or any mcp-compatible client at the server. daily use happens inside your ai client, not a terminal.
03 · honesty
this is alpha.
ai-continuum is alpha software: local-first, free to try, and no billing gates. expect sharp edges and moving parts. the safety net is the product itself — receipts on every write, and undo on everything. if that trade sounds good, you're the user we're building for.