Meta's personal AI agent, expanding through an early-access program — a video-chat avatar, Mac computer control, a dedicated email address, and a growing set of connectors that let it act across your apps and accounts.
Last verified · 2026-09-29 · by Moe Ameen
Meta Muse is the personal AI agent from Meta Superintelligence Labs, launched September 8, 2026. Where a chatbot answers, Muse acts — reading email, booking travel, filling forms, shopping, scheduling, negotiating, and paying with your card, working long-running tasks in the background and turning goals into step-by-step plans. It runs on Muse Spark, Meta's most capable model, and is available US-first on the web at muse.ai, iOS, Android, and inside WhatsApp. Chief AI officer Alexandr Wang frames it as a "second mind" that builds a plan, sends the email, makes the call, and finds the funding while you focus elsewhere.
On September 25, 2026, Meta opened an invite-style early-access program for a wave of advanced features and began promoting creator and power-user use cases. The reported additions: a digital avatar you can video-chat with (a "Muse Realtime Avatar" model), desktop control on the Mac that operates your apps and keeps working after you walk away, a dedicated email address for the agent so you can forward messages or add it to a thread, voice control through Meta's AI glasses, and an expanded roster of shopping partners and developer connectors (Meta said more than 1,500 connector apps were submitted in under a week). Sign-up is conversational — you ask Muse to add you — and several features are described as "coming months," so treat them as previews and confirm specifics on Meta's pages.
The way to read all of this as a creator is by what it acts on versus what it makes. Each new feature extends Muse's reach — more apps it can drive, more services it can touch, more ways to talk to it — but it is still a personal and operations agent, not a content producer. It drafts text and completes tasks; it renders no video, images, or graphics, holds no brand voice you configure once, sizes nothing per platform, and publishes nothing to your channels. That boundary is stable across every early-access feature, and it's exactly the seam where a content engine takes over.
Pricing is a free tier for most needs plus paid plans reported at $20 and $100 per month; the early-access features are extensions of the same agent, not a separate creator product.
The most useful way to pair Meta Muse's new agentic features with content is to treat Muse as the intake and research layer and Kompozy as the production line. Muse's advanced early-access wave is built for gathering and acting: a dedicated email address you can forward a transcript, a newsletter, or a competitor's post to; desktop control and connectors that can pull information and drive apps; planning that turns a goal into a brief. That's a genuinely handy front end for a content operation — it can assemble the raw material and the outline. What it can't do is anything downstream of the words: it renders no [Persona Short](/glossary/persona-shorts), holds no face-locked identity across a week of posts, lays out no brand-exact [Carousel](/glossary/hyperframes), and pushes nothing to your platforms. That's not a gap in Muse; it's the start of Kompozy's job.
Concretely, run it as a handoff. Forward your sources to Muse, let it summarize and draft an angle, then bring that brief into [Kompozy](/) as the seed for a content unit. Kompozy fans it into the formats a personal agent can't make: captioned [Persona Shorts](/glossary/persona-shorts) and HeyGen avatar video with one consistent face, [Persona Frames](/glossary/persona-frames) composited into brand-exact templates, Carousels and Quote Graphics, Photo Posts, per-platform Text Posts, a Blog Article, and an Email Newsletter — all held to your voice by a [Persona Brief](/glossary/persona-brief) and banned-word filters, each rendered as finished media rather than a description of it. Then [Autopilot](/glossary/autopilot) schedules and publishes the whole run across the eight social platforms plus blog and email from one queue, behind a per-post review. Muse's features gather and plan; Kompozy generates and ships — and notably, the "AI avatar" Muse gives you is one you talk to, while the one Kompozy gives you is one that presents your message to your audience as published video. For the base agent, see [Meta Muse](/ai-tools/meta-muse); for the Instagram-account angle, [Muse for Instagram creators](/ai-tools/muse-for-instagram-creators).
On September 25, 2026, Meta opened early access to advanced Muse features: a digital avatar you can video-chat with, desktop control on the Mac that operates apps for you, a dedicated email address for the agent, voice control via Meta's AI glasses, and expanded shopping partners and connectors. Several are described as arriving in "coming months," so treat them as early-access previews.
No. Muse drafts text and completes tasks, and its new features extend what it can act on — but it generates no video, images, or graphics, holds no brand voice, and publishes nothing to your channels. It is a personal and operations agent. To turn its briefs into finished, on-brand posts and publish them across platforms, you use a content engine like Kompozy.
Treat Muse as the intake and research layer. Forward sources to its dedicated email address or let its connectors gather material, have it summarize and draft an angle, then hand that brief to Kompozy, which renders the video, carousels, images, blog, and newsletter and schedules them across platforms. Muse gathers and plans; Kompozy generates and ships.
No. The Muse avatar is a face you video-chat with — a conversational interface for the agent. An AI presenter that fronts your content in finished, published Reels and Shorts is a content-generation feature; Kompozy provides that through Persona Shorts and Persona Frames with a consistent face across every clip.