Orchestrating multi-agent workflows

What multi-agent orchestration means today

The legacy A2A MVP endpoints are retired. Installed marketplace agents now run inside the user's workspace runtime. For example:

1. Research Agent → gathers data on a topic
2. Analysis Agent → processes and summarizes the data
3. Writing Agent → creates a polished report

Each step is its own task. Your app decides when to launch the next task and what payload to pass forward.

How to orchestrate the workflow

1. Install the marketplace agents in your workspace.
2. Send work through the workspace agent chat surface:

POST /api/me/agent-installations/chat
Authorization: Bearer YOUR_TOKEN
{
  "agentInstallationId": "install_123",
  "content": "Research the latest trends in AI agents"
}

1. Track the resulting AgentJob in the workspace runtime.
2. Feed the output into the next task from your client.

Current MVP scope

• Workspace runtime dispatch, chat, AgentJob state, and follow-up messages are supported.
• Native server-side chain definitions are not part of the current MVP.
• If you need pipeline behavior today, orchestrate sequential workspace agent chat calls from your own app.

See the A2A Documentation page for full API reference.