Build an agent cluster

Group two or more agents so they share user identity, memory, and threads. The classic case: a chat agent plus a BGO agent that "feel like one" from the user's perspective.

The goal

A cluster of agents whose memory writes are visible to every member, whose threads are shared, and whose messages carry attribution (authorAgentId) so the UI can show "Wally Chat said X, Wally BGO said Y" within one conversation.

Steps

1. Create the agents separately.

   Two agents: wally-chat (model: a fast chat model; tools: chat-friendly) and wally-bgo (model: a stronger reasoning model; tools: code-runner, scrapers).

2. Open the clusters page.

   Sidebar -> Agent clusters -> "New cluster".

3. Configure.

   • Name: Wally.
   • Slug: wally.
   • Members: pick wally-chat and wally-bgo. The dropdown only shows agents not already in another cluster.

   Save.

4. Use the cluster from chat.

   In a chat with wally-chat, prompt: "Render the slides into a PDF and reply when ready." The chat agent calls spawn_bgo targeting the wally-bgo member; the BGO runs; its eventual message lands in the same thread, attributed to wally-bgo.

Verify

• recall from wally-bgo returns memories written by wally-chat (cluster scope).
• The thread shows messages from both agents with their respective avatars.
• An agent in a different cluster cannot read this cluster's threads (IDOR check rejects).

Memory before vs after the cluster

Memory written before a cluster was formed stays scoped to the original agent. Cluster scope kicks in on writes that happen after the link.

Next steps

• Extract long-term memory using the cluster scope.
• Spawn a BGO for the cross-member dispatch flow.