Skip to content
NagentNagent
Log inSign upHire your AI team
Atlas/Muse (Meta)
Vendor dossier / Autonomous watchlist

Muse (Meta)

Meta’s consumer personal AI agent; shared enterprise context and growth-team governance require separate evidence.

REVIEWED 23 SEPTEMBER 2026Marketing / Sales / CXNO HANDS-ON BENCHMARK

What it is

Official-source notesAutonomous watchlistwatchlist

Meta announced Muse as a personal AI agent on 8 September 2026. Its product page shows a dedicated virtual machine and browser, connected personal apps, task memory, approval prompts for email and purchases, and an activity trail. Muse is distinct from the Muse Spark model family and Muse Image. The public material does not describe a multi-person company workspace. [V1] [V2]

Documented means the statement is present in the cited material. Capability effectiveness, customer outcomes and security assurances have not been independently audited here.

Promoted use cases

  • Arrange personal travel and appointments
  • Send emails and handle shopping after approval
  • Coordinate personal goals and tasks across connected apps

General agents matter because they change expectations about what an assistant can do. Their flexibility is also why an impressive personal demonstration does not establish readiness for shared enterprise work. This category is a watchlist: product identity, team access, permission boundaries and operational support need clearer evidence before treating an entry as a direct growth-platform substitute.

Pricing and buying model

Published commercial evidence

Free core; paid plans

Meta describes free access and optional subscriptions, but the official pages reviewed did not state a reliable plan-by-plan price. This is a US consumer offer, not an enterprise growth quote.

Check current commercial source ↗

Budget for implementation, model and tool usage, data, human review and ongoing support. Credit units and outcome definitions differ between vendors. A missing numeric rate is marked unverified rather than replaced with an old third-party estimate.

Collaboration and governance

Meta states that a separate Sentinel agent screens external actions and that sensitive actions request a person’s approval. These are vendor-described protections, not an independently audited enterprise governance programme. A later Confidential VM is a planned feature.

The relevant unit of control is permission to interact with a connected application. Inspect who can propose, approve, execute, interrupt and audit that action. Shared seats, shared content and a shared live agent session are different capabilities; require a demonstration of the one your process needs.

Evaluate memory correction, permission revocation and release control for updated instructions. An improvement loop should preserve the original evidence, proposed change, test results and named approver. A safety or security badge alone cannot establish those workflow properties.

Review and customer evidence

A WIRED hands-on article raises privacy and permission-transparency concerns. This is one reviewer’s reported experience, not aggregate customer sentiment or a controlled benchmark. See the independent-review link below.

Before relying on a testimonial, confirm the exact product, version, package, workload and baseline. Ask a relevant customer about setup effort, failed cases, ongoing manual work and support after launch. This dossier does not convert customer logos or vendor-hosted awards into independent proof.

Focused feature documentation

Focused documentation check · Sep 2026

Muse is a live personal agent, not an unresolved model name

Meta announced Muse on 8 September 2026 as a consumer personal agent available in the United States. Its official product page shows a persistent virtual machine and browser, connected personal apps, task execution, memory, sensitive-action approval prompts and an audit trail. Meta describes a separate Sentinel agent that screens external actions. These are published design claims; the illustration is not a product test.

Muse the personal agent is distinct from the Muse Spark models and Muse Image. The official pages reviewed do not establish shared organizational task ownership, a growth-team administrator console or a comparable enterprise price. It is best used as an adjacent agent design reference rather than a direct Dust or Nagent replacement.

Primary sources: Meta Muse product page ↗ · Meta launch announcement ↗

Published documentation and public illustrations; account behaviour was not independently tested.

Visual source

Public Muse (Meta) source page captured 22 September 2026
Muse (Meta) · Public page captured 22 Sept 2026. Vendor presentation, not an authenticated product test. Open source ↗

Fit, limitations and proof requests

Editorial assessment

Muse is an adjacent consumer agent, now a verified product rather than an unresolved name. Its task planning and action controls are relevant to the category, but a personal agent is not a direct substitute for a governed growth team with shared organizational ownership.

Priority question: Can multiple authorized people govern the same agent task, and does Meta offer a separately documented enterprise deployment for growth work?

The evaluation owner should be an accountable pilot owner. Use useful completed work without unapproved side effects as the business target. Judge the solution in its own stack role: a cloud runtime, a specialist production tool and a managed AI team can be complementary purchases.

Use-case evaluation plan

A bounded research-and-artifact task

Choose a task using non-sensitive public inputs and require a sourced result with a clear completion condition. Include an ambiguous instruction and observe whether the agent asks an appropriate question or invents a direction. Keep the task limited enough that its output can be reviewed completely.

Measure: source accuracy, task completion and human correction effort.

Handoff between two humans

Where a team capability exists, ask a second authorized person to inspect progress and take ownership. Check whether account credentials, private context and pending actions remain correctly scoped. If the feature is undocumented, mark the result unknown rather than simulating multiplayer capability through shared passwords.

Measure: ownership clarity, access boundaries and continuity of task state.

Memory and unattended execution

Inspect how the agent stores useful facts and schedules repeat work. Correct an erroneous memory and revoke an integration before the next run. Test whether the change takes effect and whether a human can stop the process without losing the evidence needed to understand what happened.

Measure: memory correction, revocation behaviour and safe task interruption.

Detailed comparison

Read the detailed Nagent vs Muse (Meta) guide for operating models, shared context, governance, cost, evidence gaps and a staged pilot.

Nagent vs Muse (Meta)

Sources and evidence register

Nagent · Multiplayer AI for growth teamsResearch: 21–29 September 2026 · Public-source analysis, no hands-on benchmark · Sources & method