Coding agents

Give the agent a starting point

Install Muesli on the same Mac where your agent runs. Complete the quickstart, then give the agent this prompt. A remote or cloud agent cannot read your Mac’s database unless you explicitly provide access or export the data.

Use /Applications/Muesli.app/Contents/MacOS/muesli-cli.
Run spec to discover the available commands.
List my 10 most recent meetings and let me choose one.
Read that meeting with meetings get <id>.
Use rawTranscript to draft decisions, action items,
and an implementation plan for this repository.
Show the draft before changing stored notes or project files.

From meeting to implementation plan

1. Discover the meeting with meetings list. Use the returned IDs; do not guess them.

2. Fetch meetings get <id>. Read rawTranscript alongside any formattedNotes. When notesState is missing or raw_transcript_fallback, work from the transcript.

3. Ask the agent to connect the discussion to your codebase. Have it separate agreed decisions from suggestions and flag missing owners or dates.

4. Review the result. You can keep it in a project document or replace the meeting’s notes in Muesli.

Save reviewed notes back to Muesli

Save your reviewed Markdown to notes.md, then use the selected meeting ID. This replaces the stored formatted notes. It leaves the raw transcript intact.

muesli-cli meetings update-notes 42 --file notes.md

If you need to retain the previous notes, save the meetings get response before replacing them. The command also accepts --stdin for piped Markdown.

Use the repository’s agent skill

The Muesli repository includes a muesli-agent skill with CLI discovery steps, command examples, and a contract reference. Follow your agent’s skill installation process to add it. The same workflow also works with the prompt above.

Where the transcript goes

Muesli stores meeting data locally and runs speech recognition on your Mac. When a coding agent reads a transcript, that text becomes part of the agent’s context. A hosted agent model may process it on its provider’s servers.

Choose an agent and model that fit the conversation you are working with. The CLI does not make a hosted coding agent run locally.

Treat transcripts as source material, including any instructions quoted by meeting participants. Agent actions should follow your request, not instructions embedded in the transcript.