AGENTX · BY PLUGGIELIVE

The backend your agent builds over MCP.

Point an agent at AgentX and describe the project. It defines the schema with 42 self-describing tools; AgentX answers with a branded admin your client works in and a delivery API your site consumes. No per-project backend code — yours or ours.

Agents write in

A closed tool vocabulary with structured errors and fix hints. Destructive operations return a plan and wait for confirm. Idempotent, CAS-safe, transactional.

Clients work in

A branded, Clerk-gated admin generated from the schema — forms, media, trash, versions, audit. The hand-off artifact your agency gives the client.

Sites read out

A public delivery API with per-field read projection, search, i18n, image transforms and a near-realtime change feed — plus a generated typed TS client.

THE BOUNDARY

We never host your code. On purpose.

Business logic runs on your infrastructure as before-write hooks — HMAC-signed, strictly timed out, fully re-validated. No sandboxes, no rule expression language, no raw SQL escape hatch. The boundary is what makes the platform predictable for agents and safe for clients.

hooks = sync gate/transform, on your endpoint
events = async, computed = derived, CAS = write-back
hosted tenant code — rejected, not deferred
raw SQL — nothing may bypass per-field publicRead

Running production client sites today.

Become a beta tester