Create a new named persistent agent. The agent stays alive in memory with full conversation history until killed.

Send a message to a persistent agent and receive its response. The agent remembers all prior conversation. It can use tools (read files, search, run commands) to do real work.

Send a message to all agents (or a filtered subset by tags). Optionally collect all responses.

Get an overview of the agent pool, or detailed status for a specific agent.

Write a key-value entry to the shared memory bus. All agents and Master Control can read it.

Read entries from the shared memory bus. Can read a specific key, all keys with a prefix, or all entries.

Export memory bus entries matching a prefix to a JSON file on disk. Use for pipeline recovery, monitoring, and reports.

Terminate a named agent. Its conversation history and context are permanently destroyed.

Clear an agent's conversation history and token counter, but keep the agent alive with its configuration intact.

Update an agent's configuration at runtime (model, history limit, token limit, system prompt, tags).