Muse Code exits beta with developer-preview SDK
Muse Code left beta and released a developer-preview SDK for embedding custom agents, tools, progress streams, and resumable sessions. Its new workflows can split work among focused agents that pass intermediate context between sessions.

TL;DR
- Muse Code has exited beta, as finkd's announcement says, continuing Meta's terminal-agent project powered by Muse Spark 1.2, per Meta's original launch post.
- The developer-preview SDK can build custom Muse agents, embed them in applications, attach custom tools, stream progress, and resume sessions, according to finkd's SDK announcement.
- Separate coding sessions can now exchange context and state directly when their work intersects, finkd's session-messaging post says.
- Workflows can fan a task out to focused agents, carry intermediate output between stages, then produce one result, per finkd's workflow post.
- Monthly subscriptions are rolling out, finkd's subscription notice says; RuntimeWire's launch report puts the entry plan at $5 a month.
The Muse installer verifies a server-provided SHA-256 checksum and can update shell PATH settings, with MUSE_NO_MODIFY_PATH available to prevent that change. Ollama says in its integration post that the Muse harness already runs with local and cloud models through ollama launch muse.
Beta exit
Meta's August 5 beta announcement described Muse Code as a terminal agent with persistent async subagents for planning, code changes, and validation across large repositories. The new announcement takes that product out of beta.
SDK developer preview
The finkd SDK announcement names five capabilities in the preview:
- Build agents on top of Muse Code.
- Embed those agents in an application.
- Connect custom tools.
- Stream progress.
- Resume a session later.
Inter-session messaging
Inter-session messaging lets sessions share context and state when their work affects each other, removing the manual copy-paste handoff described in finkd's post. The release defines this as direct coordination between ongoing sessions, not merely more subagents within one task.
Workflows
The rollout thread describes the target as complex work divided among multiple teams of subagents. The stages that finkd's workflow post specifies are:
- Break a task across focused agents.
- Carry intermediate work between stages.
- Return one result at the end.
Ollama harness support
Ollama says the Muse Code harness works out of the box on its local and cloud model surfaces through ollama launch muse.
Subscriptions and Model API availability
Monthly subscriptions are rolling out, finkd's notice says, and RuntimeWire's report lists a $5 monthly starting tier. Meta's Model API page separately lists country-based availability limits for API access.