Let Muse be your gardener
A bonsai doesn’t need you all day. It needs you at the right moment: when a branch has to be wired, when the soil has gone dry. Miss that moment and nothing breaks at once. The tree just grows the wrong way for a while.
Our board works like that now. Agents do most of the growing. A writer codes, a reviewer reads the PR, a deployer ships. Every so often one of them stops and waits for a person. When it needs you to choose, say between two approaches, Bonsai puts that decision in the Inbox under Need you, together with any run that failed.
The Inbox has one weakness. It only helps when you open it. The place you actually want to be told is your phone, and our own app isn’t there yet.
A gardener that walks by
Muse is Meta’s assistant, and two things make it a good gardener for Bonsai. It runs on its own machine in the cloud, so it can do things on a schedule while your phone is in your pocket. And it connects to remote MCP servers as custom connectors. Bonsai is an MCP server, so we connected the two.
Setting it up takes two messages to Muse. The guide has both, word for word:
- Connect Bonsai as a custom connector. You sign in to Bonsai and choose what Muse may see, the same way you connect Claude.
- Give Muse a recurring task: every so often, ask Bonsai what needs you.
From then on Muse checks your Inbox at the interval you picked. When nothing needs you, it stays quiet. When a decision is waiting, you get one message: the question, the option the agent recommends, and a link that opens that card in Bonsai. If you don’t act on it, Muse asks again later. That second ask is the part we like best. A notification you swipe away is gone for good. A gardener comes back.
Only what needs you
By default Muse brings you Need you and nothing else: decisions waiting in an issue, and runs that failed and need dispatching again. You should know and Completed stay in Bonsai for when you sit down with it. A PR to review or a staging build to check isn’t an Inbox item: those issues sit in Your turn on the board, and the guide has one more line for the task if you want Muse to watch that column too. A reminder that repeats is only kind when it’s rare, so we started narrow.
Talking back
Muse doesn’t only read. Over the same connection it can:
- Capture what you say into Bonsai, word for word, as a note in
0_Inbox/or a line in today’s daily note: “Tell Bonsai: the pricing page needs a FAQ.” - Answer “what’s being worked on right now?” Bonsai’s issue list now knows which issues have a run going at this minute, on which machine and in which role.
- Find “issues I own that changed since yesterday”, or anything with “login” in the title.
- Move an issue along its workflow when you ask it to.
When Muse creates or edits a note, the change waits under You should know like any other agent’s, with a short summary. Decision cards you still answer in Bonsai, one tap from Muse’s link. Once the last open decision on an issue is answered, the issue moves on and the next run starts on your machine, and we want that to happen in one place you can see.
What Muse can’t do
Muse sees only what you granted when you connected it, and a read-only grant can’t write. Text inside a note is data: it never changes what Muse is allowed to do. Bonsai’s server still doesn’t run a model. Muse thinks on Meta’s side, your agents think on your own runtime, and Bonsai is the garden both of them tend.