--- title: "Three reasons I skipped Meta's Muse and one trigger to revisit" canonical: https://dxdev.com/everyday-ai/three-questions-before-a-popular-ai-tool/ datePublished: 2026-10-15 --- My partner kept telling me to use Muse, Meta's personal agent. It connects to Gmail, a calendar, Facebook and Instagram, and it can run tasks in a web browser. I wanted to know whether it was worth building into my flows. I told the AI assistant I already use for the business: "maybe we should start by seeing if I can get you to talk to it." That first step presumed there was a way for the two to talk. ## What the first round of research found I had my assistant send a sub-agent to look into it. The first job was making sure it was the right product, because an older video-hosting site has a similar name. The login on this one redirects through Facebook's auth, and third-party coverage also says Muse is Meta's. The sub-agent did not log in or sign up anywhere. On the question I cared about, it found no public API, no MCP server and no webhooks. My assistant passed that along and added that this was only what the search turned up, not proof that no such channel exists. It also named one possible route: texting Muse over WhatsApp. It offered a test. If Muse had a WhatsApp number, it could send a message and see whether the reply came back in a form it could read, which would show in about 15 minutes. I did not take the test. I replied "research, see what people are doing." That was the second round. ## Where I was wrong The second round found no reports of a WhatsApp or text channel at all. My assistant corrected itself: the test it had suggested had nothing to test, so drop it. The only route the first round had named was gone, and it took a second research round to find that out. The picture from that round was thin. Third-party write-ups put the launch at September 8. Coverage was mostly launch write-ups and vendor-style blogs. Reviewers described errands: filling out a field-trip form found in an email, drafting a refund follow-up, turning a saved Instagram recipe into a grocery list. Email triage and calendar were reported to work, and harder tasks were hit and miss. The sub-agent found no founder or developer describing day-to-day use. The only first-person reactions were on Hacker News, and in the first 100k of roughly 366k characters it read, no commenter said they use it. One source said you can give Muse the web address of a tool server and it will call that tool. So Muse could use tools, but we could not drive Muse. A Muse-specific setup guide said [installing Claude Code](/blog/2026-05-04_gmail-identity/) or Codex inside its cloud computer isn't supported. ## What it came to 1. **Can our assistant reach it?** No way in turned up. There is no public API and no developer portal, and the sub-agent saw no roadmap for one as of mid-September. 2. **Does it do something we don't already do?** Not that we could find. We already have mail, calendar, browser automation and memory, and they run under rules we set. Muse would add a second memory store we don't control. 3. **Where does our information go?** To Meta. Meta says Muse data stays separate from ads and that you can opt out of training. That is Meta's claim and I haven't checked it. I would keep customer data from our platform out of it entirely. Two details stayed open. One source says "SOUL" and "MEMORY" are editable [files behind two cards](/blog/2026-10-11_speaker-provenance-words-store/) in the app. Neither Meta nor anyone else confirms you can export or import them. If you could, that would be the most useful piece, because we could seed it from our own notes. ## Where it ended up I told my assistant to store it as an option in our list of outside tools and close the session. The whole thing took 33 minutes. The entry says Muse was reviewed and skipped because it has no API, duplicates the tools we already run, and keeps its data with Meta. It says to revisit if Meta ships a developer API, and it notes the Muse Spark model as a possible cheap subagent model, which would be metered spend and needs my go. The entry has three reasons to skip it (no API, duplicate tooling, data that sits with Meta) and one trigger to look again: a developer API from Meta.