APIs & SDKs

Which surface do I need?

Surface	Use it to	Docs
Studio API	Manage things: agents, knowledge bases, skills, tools, connections, roles, audit, traces.	Studio APIs
Public API	Run published agents from any backend over plain REST with an API key: sync, async or streamed.	Public API, API keys
MCP server	Let Claude, Cursor or any MCP client find, run and configure your agents, signed in with OAuth.	MCP server
Sandbox / runtime API	The session gateway the Studio chat uses: start a turn, stream the answer, answer an input request, cancel.	Sandbox SDK

Loading diagram...

A live, browsable OpenAPI reference is served by every Studio deployment:

GET  {STUDIO_URL}/api/rpc/api-reference
GET  {STUDIO_URL}/api/rpc/api-reference/spec.json

It is generated from the same schemas the server validates against, so it is never out of date.

Caller	Credential
Browser / first-party app	Session cookie from sign-in
Server-to-server	Studio API key (`studio_...`), scoped to one organization. Acts as the person who created it.
MCP client	OAuth 2.1 sign-in, scoped to one organization and the permissions you consent to

Every call is checked twice: your role decides what you may create, and a per-object permission decides what you may reach.

IDs are UUID v7. Time-ordered, so they also sort chronologically.
Lists are cursor-paginated. Pass the last id as cursor; limit is 1 to 100 (default 50).
Nothing is hard-deleted. Deletes are soft; the object disappears from reads.
Secrets never come back. API keys, OAuth tokens and vault contents are write-only.
Sensitive calls are audited, including the ones that are refused.

Public API: run agents over REST. Start here for most integrations.
API keys: create, scope and revoke keys.
MCP server: connect AI clients to Studio.
Studio APIs: the management surface, module by module.
Sandbox SDK: the runtime surface for embedding agents.
Self-hosting and operations: signup modes, background jobs, configuration and rate limits for operators.