A local-first AI workforce that runs on your own machine. A Chief of Staff reads the request, decides whether it needs specialists, assigns the work with acceptance criteria, and comes back with one answer and a readiness verdict.
Nineteen specialists each hold their own identity, expertise and memory, and they're allowed to disagree: if Security finds a critical flaw or QA finds a failing test, the Chief of Staff reports NOT READY rather than smoothing it over. Anything that reaches outside the machine, spends money, or can't be undone stops in an Approval Inbox with the exact payload shown before it runs.
Memory carries provenance — where a fact came from, when, and how confident it is — so speculation is never stored as fact. The audit trail is append-only and hash-chained, so a modified or deleted record breaks the chain and is detectable rather than silent. Everything except model inference stays on the machine, running in Docker across an API, a worker, an MCP server and a web console.
There is no hosted instance to link. Meridian is built to run on your own machine, behind your own sign-in — a public demo would defeat the point.