Skip to main content
Pushing knowledge is one half of the API. The other half authors the agents themselves: everything the dashboard configures is config-as-data you can write over the API — no field is dashboard-only.

​
What you can author

  • Agents — provision a new agent from a starter, then tune its widget settings.
  • Catalog — the services an agent offers, the fit criteria it qualifies against, and the voice it speaks in.
  • Flows — the conversation itself: its phases and the answers each one captures.
  • Drafting — let pepline propose that catalog from what your agents already know, and accept what fits.
The same surface serves three callers: pepline’s own concierge onboarding, the pepline MCP server, and your own tooling.

​
Conventions that hold everywhere

  • Declarative activation. Many resources take an agents array of slugs that is a full replacement set — the resource ends up offered by (or active for) exactly the agents you list. Omit it to leave activations untouched; send [] to deactivate everywhere.
  • Archive, never delete. Nothing is destroyed. Archiving removes a resource from an agent’s world while keeping its history; restoring brings it back. (DELETE on a source answers 405 and points you at the archive verb.)
  • Flows trial-compile. Every write that touches a flow is compiled in the same transaction before it commits. An illegal change — over the slot budget, a reserved key, moving the closing phase — rolls back with 422 flow_invalid. You cannot author a broken agent.
Everything here is authenticated the same way as the rest of the API — see Authentication.