Use Thrum with Muse

With the Thrum plugin installed in Muse, you can ask your coordinator to put an agent on Muse. Your coordinator is the agent you talk to, which starts the other agents, hands out work and reports back to you. Muse is an AI coding tool that you run in a terminal with the muse command. A Muse agent shares the team’s messages and starts with a briefing, a summary of who it is and where the project stands, like your Claude Code agents.

Muse has its own plugin system and marketplace. It does not use the Claude Code marketplace. Muse installs the same Thrum plugin with its own commands. The quickstart starts your coordinator on Claude Code. This plugin matters for the agents the coordinator creates on Muse.

What the plugin gives you

• Thrum’s skills inside Muse. A skill is a set of instructions an agent can load, such as the playbook for a role. Muse lists the plugin as thrum. To run a skill by hand, type /skills use plugin:thrum:<skill>.
• A briefing at the start of every Muse session. Muse has no start-of-session hook that Thrum relies on. Instead, Thrum types /thrum-prime into the agent’s session for you when it starts the agent. If you start a Muse session yourself, type /thrum-prime there.

Before you start

• Thrum set up in your repository, with a running coordinator. The quickstart starts one on Claude Code. That is the recommended way to begin, and the steps on this page assume it.
• A Muse account and the muse binary. Muse updates itself when it starts.
• You signed in to Muse (muse login). You do this yourself. Your coordinator does not sign in for you.
• The thrum binary on your PATH.

Add Muse agents through your coordinator

Caution

Thrum starts Muse agents with Muse’s sandbox turned off (--disable-sandbox), because with it on the agent cannot register with Thrum. Thrum always starts them this way. Nothing in Muse then limits an agent to your repository: it can read and change whatever your user account can. Give a Muse agent only work you would trust with that access. --trust-workspace skips Muse’s question about trusting the folder. Muse can still ask before some commands, such as several joined into one.

1. Ask your coordinator (in its terminal, or in My Inbox, your inbox in Thrum’s web UI):

   Install the Thrum plugin for Muse in this repository.

   Muse ships a plugin’s hooks switched off until you approve the plugin. A hook is a small script that a tool runs at a set moment. Your coordinator runs the approval, or asks you to. If it asks you, run muse plugins approve thrum in a terminal in your repository.

2. Check that the plugin is installed. Ask your coordinator: Is the Thrum plugin installed and approved in Muse?

3. Ask your coordinator for a Muse agent. Say “on Muse” in your message. The runtime name is muse. A runtime is the AI coding tool an agent runs in, such as Claude Code or Codex. This is an example conversation. Your coordinator’s wording will differ.

   You:
     Create an agent on Muse to review the README. Give it its own worktree.
   
   Coordinator:
     Created reviewer_readme in a new worktree and sent it the task. I'll tell you when it replies.

   The New Agent dialog in the web UI has no runtime field and always creates Claude Code agents. For Muse, ask the coordinator.

4. Check it worked. On the Agents grid, a card for the new agent appears. Open that card, then its Terminal tab, which shows the Muse agent’s own screen. You see /thrum-prime run, then one line that names the agent:

   @reviewer_readme primed (reviewer/main). Reviewing README.md. Standing by.

   Muse can ask “Do you trust this workspace?” Thrum starts Muse with --trust-workspace, so the question does not appear. If it appears anyway, Thrum answers it for you.

   If the primed line does not appear, ask your coordinator: Is the Thrum plugin approved in Muse?

If your coordinator will run in Muse

Skip this section if your coordinator runs in Claude Code. If Muse will be your first agent, there is no coordinator to ask yet, so install the plugin yourself. Run all these commands as the same OS user that runs your agents, in a terminal where you can type answers. Step 3 fails from a script or a background job.

1. Sign in to Muse, if you have not yet.

   muse login

2. Add the Thrum marketplace. thrum-team/thrum-pro is where the plugin is distributed from. It holds the plugin’s text files, not the Thrum program. Free and Pro both use this same plugin. Thrum is published by THRUM LLC and built by Leon Letto (About). In the command, thrum is a label you choose for this source, and thrum-team/thrum-pro is where it comes from.

   muse plugins marketplace add thrum thrum-team/thrum-pro

3. Install the plugin.

   muse plugins install thrum@thrum

4. Approve the plugin’s current capabilities.

   muse plugins approve thrum

   To see the current status, run muse plugins inspect thrum --json.

5. Check it. Run muse plugins list. It shows thrum as installed.

Update the plugin

Refresh the marketplace, remove the plugin, install it again and approve it again:

muse plugins marketplace update thrum
muse plugins remove thrum
muse plugins install thrum@thrum
muse plugins approve thrum
muse plugins list

Then run /skills reload in each running Muse session to load the new skills.

Tips

Tip

To update the plugin, ask your coordinator: Update the Thrum plugin in Muse, then reload the skills in every running Muse agent. Muse asks you to approve the plugin again after an update; your coordinator runs that approval, or asks you to. Each running Muse session needs /skills reload after an update, with no restart, and your coordinator can type it into each agent’s terminal. A change to the hooks needs a restart of the session.

Muse’s own muse plugins update thrum can fail or leave the old version in place, so the update removes the plugin and installs it again. If your coordinator runs in Muse, there is no other coordinator to ask: use the steps under Update the plugin.

Note

After a restart, Thrum types /thrum-prime again, so the agent gets its briefing back. Mix runtimes and models has more on how runtimes differ.

Next steps

• Mix runtimes and models: choose the tool and model for each task.
• Messaging between agents: how the coordinator and a Muse agent talk.
• Use Thrum with Antigravity: the same idea for Antigravity.