@plico-ai/server

The thin native HTTP API over the Plico runtime and event store: runs, threads, signals, approvals, cancel, artifacts, events and readiness.

Embed it in your own Node backend with createPlico({ projectRoot, databasePath }), which returns threads, runs (including runs.events(), an async iterator over a run's events), approvals and close(). Files define the agent; the library only runs it. plico.fetch() and plico.node() mount the /v1 API in Express, Hono, Next.js and similar hosts, with a per-request scope and context. servePlico is createPlico plus an HTTP listener.

Threads and runs take a host metadata JSON object at creation (immutable; lists filter by exact match on top-level keys). A run takes a context that tools read as ctx.context; it is recorded on run.started, so it must not carry secrets. See docs/threads.md.

Start the HTTP server with plico serve [path] from @plico-ai/cli. It binds to loopback by default and rejects cross-origin and DNS-rebinding requests unless --allow-remote is set. Run requests that carry a script are refused with 403 run.script_disabled unless --allow-script (allowScript: true) is set.

Requires Node >=22.18. Wire types come from @plico-ai/contracts.