A free consumer agent or a private personal agent?
By TallisReviewed Public-source comparison
Compare the work. Then choose the tool.
Tallis’s column summarises its current product description. Competitor capabilities are attributed to the sources cited, not independently tested. “Not verified” does not mean unavailable.
Everyday consumer errands for individuals: email, travel, forms, purchases and lists. [1][2][3][4][5][6][7][8][9]
How you work
Tell it the job in a chat, then close the app. [T]
The Muse app and WhatsApp. Available only in the U.S. and Canada as reported on 2026-09-25, so not in the EU or UK. [1][2][3][4][5][6][7][8][9]
Work to hand over
It browses, researches and drafts on its own computer. [T]
Errands and purchases run by an agent that browses as you inside a dedicated Meta-hosted secure VM. [1][2][3][4][5][6][7][8][9]
Control over actions
Its own computer, separate from everyone else's. You can stop a running job, and model costs are capped. [T]
Meta's own announcement describes a separate Sentinel agent on the same machine that must approve anything reaching the internet, and says Muse checks with the person before sensitive actions such as sending email or making a purchase. What the person is shown, what happens if they never answer, and what happens when an outcome is unknown are not stated. Meta's terms were not re-verified. [1][2][3][4][5][6][7][8][9]
Access and cost
Waitlist. Confirm launch pricing and workflow availability before subscribing. [T]
Free; Power $20/month; Maximum $100/month. Paid tiers buy more usage, not more features. Available only in the U.S. and Canada as reported on 2026-09-25; not offered in the EU or UK. [1][2][3][4][5][6][7][8][9]
Start with your next useful task.
Consider Muse if you want free consumer errands inside Meta's apps and are comfortable with its data terms. Consider Tallis if you want a private agent for recurring operator work whose memory, files and browser never feed another company's models, and which is not tied to one company's model: Muse runs Meta's, and a company that sells a model does not route your work to a rival's. Read Meta's current settings and terms yourself; this page cites press reports and Meta pages checked between 2026-09-10 and 2026-09-30.
This is our assessment of fit. We have not run a head-to-head performance or reliability test.
A useful test beats a feature count.
Suggested evaluation steps, not a test we have performed.
Run one errand that touches a merchant account, then check what the agent retained and where it lives.
Find the training opt-out and the memory controls without help, and time how long that takes.
Ask for the same recurring brief three weeks in a row and compare what each remembers.
Questions worth asking.
Is a free agent private?
Not by default, according to Wired's hands-on review of 2026-09-20: Muse users are automatically opted in to having their interactions with the agent used for AI model training, and the opt-out is a toggle in settings. Meta says Muse does not share conversations or data with its ads systems. Meta's own security description, an isolated machine and a separate approving guard, is real and does not change the training default; isolation is about who else can reach your agent, not about what the operator may do with what passes through it. Tallis does not use your conversations to train AI models. Verify Meta's current terms before deciding; this page cites dated press and Meta pages, not Meta's terms. [1][2][3][4][5][6][7][8][9]
How current is this comparison?
We checked the cited material on 2026-09-30. Prices, availability and features may change. Confirm the terms and the exact workflow with the provider before committing.
Sources and review
Published by Tallis, one of the products compared. This is an editorial comparison of public information, not an independent product audit. We separate published claims and our assessment of fit.