Agents

An agent is an autonomous Claude Code (or Codex) session running on a cloud VM. Agents clone repos, execute tasks, open pull requests, and can spawn child agents recursively. The platform manages the agent’s full lifecycle — provisioning, credential delivery, task execution, follow-ups, sleep/wake, and termination. You interact with agents through the CLI, MCP tools, or dashboard.

​

Agent lifecycle

• Provisioning — a warm VM is claimed from the pool
• Running — the coding agent is active, working on the task
• Idle — the agent finished its turn and waits for follow-ups, events, or timeout
• Sleep — VM is suspended. Compute charges stop, state is preserved. Wake to resume.
• Wake — the same VM resumes in seconds with session state intact
• Kill — the agent session is persisted and the VM is terminated

​

Session modes

The idle behavior is configurable — agents can sleep (default), terminate immediately, or keep alive indefinitely. See the --on-idle flag on murmur spawn.

​

Slugs and identity

Every agent has a slug — a short identifier you choose at spawn time (e.g. fix-auth-bug). Slugs are DNS-label-safe (lowercase, alphanumeric, hyphens). The full agent identity combines your account and slug — e.g. alice/fix-auth-bug.

​

Recursive spawning

Agents can spawn children, and children can spawn children. This is the foundation of multi-agent orchestration. The parent receives a child_lifecycle event when children complete or fail. See the multi-agent orchestration guide.

​

Follow-ups and queue

You can send messages to running agents via murmur queue add or the queue_add MCP tool. Messages are queued and delivered when the agent is ready (between turns). The dequeue strategy controls how many queued messages are delivered per turn — all at once (default), one, or up to five.

​

Session resurrection and forking

Resurrect resumes a prior session — same slug, same branch, full conversation history. Fork branches off another agent’s session into a new independent agent. Both are flags on murmur spawn.

​

Agent personas

An agent persona is a reusable configuration that defines a system prompt, model preference, tools, and constraints. Spawn with a persona using the --agent flag on murmur spawn.

​

Backend selection

Agents run Claude Code by default. Pass a Codex model or --backend codex on murmur spawn to use Codex instead. The --reasoning-effort flag controls thinking depth.

---