Publish agents for governed in-chat work
Connect and list agents so people can select them in the Agent Hub and run them in chat.
- End users pick listed agents in the hub; invocations stay in the same conversation.
- Integration is not bundled with a self-hosting walkthrough first.
- Open APIs match the product UI: routing, runtime rules, confirmation before sensitive steps, and reviewable execution history.
Quick start
Connect your first agent in minutes. Mirrors the in-app Provider flow: register, health check, test invocation, listing.
Read guide →SDK & API
TypeScript, Python, REST. The reference expands over time; Quick start remains the fastest path to a first connected agent.
View reference →A2A protocol
How the platform reaches your agent endpoint: discovery, capabilities, requests, errors. Runtime rules and execution records keep invocations traceable and reviewable.
Read protocol →Provider developer portal
W-65 / OPT-005 & OPT-007: context_mode decision path, GitHub session-history & Agent input spec + adapter guide, and one-line degit to clone the A2A adapter template.
Open portal →15-minute minimal listing
Shortest path from a working A2A endpoint to a listing: read the contract, echo once, declare concurrency, submit. Matches in-app Provider console checks.
Read checklist →Provider upgrade assistant
Wizard for A2A 1.0.0 alignment: pre-recheck checklist, sample Agent Card JSON, and explanations for common failed_check_ids (signatures, tenant paths, ListTasks).
Open assistant →OpenClaw & Google A2A
Agent Card at /.well-known/agent-card.json, protocol_version google_a2a_v1, SendMessage wire format—zero custom shim on the OpenClaw side.
Read guide →Desktop Connector
Local channel between the platform and the user’s machine. PROTOCOL.md describes messages and pairing; installers live on the download page.