Muse Code agent runtime

OpenClaw Muse Code agent runtime plugin (registers a muse AgentHarness backed by @muse-code/sdk over the Muse Session Protocol to the Muse Code CLI)

Muse Code agent runtime (OpenClaw plugin)

External OpenClaw plugin that registers a muse agent harness backed by @muse-code/sdk and Meta's Muse Code CLI (muse serve).

Install

bash
openclaw plugins install @openclaw/muse

Install Muse Code separately and sign in once on the Gateway host with muse or muse login.

The harness is opt-in only: set agentRuntime.id: "muse" on a meta/* model or on the meta provider; auto never picks it.

See Muse Code agent runtime for auth, approvals, sessions, limitations, and live tests.

Chat switches from preparation to waiting when Muse accepts the turn, then streams assistant and tool activity through the Gateway. Once the answer is saved and Muse starts its post-answer reminder work, the chat activity indicator stops. The run remains active until Muse reports turn completion, so Stop and queued messages keep their normal behavior during background finalization. Native tool starts restore visible activity, including when they follow assistant text. finishing settles an attempt rather than the logical run: newer retry, fallback, or foreground activity restores the indicator, while older replayed events cannot replace the newer state.

Package

  • Plugin id: muse
  • Package: @openclaw/muse
  • Minimum OpenClaw host: 2026.10.2