--- title: Meta's Muse AI agent: what Zuckerberg said about... description: Meta launched Muse, a persistent personal AI agent designed for ongoing tasks like meal planning, health tracking, and project management rather than one-off... canonical: https://daily.dev/posts/meta-s-muse-ai-agent-what-zuckerberg-said-about-privacy-pricing-and-llama-4-y8zu7g1wa twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Meta's Muse AI agent: what Zuckerberg said about privacy, pricing, and Llama 4 | daily.dev og:description: Meta launched Muse, a persistent personal AI agent designed for ongoing tasks like meal planning, health tracking, and project management rather than one-off... og:url: https://daily.dev/posts/meta-s-muse-ai-agent-what-zuckerberg-said-about-privacy-pricing-and-llama-4-y8zu7g1wa og:image: https://api.daily.dev/og/posts/y8zu7g1wa.png og:image:alt: Meta's Muse AI agent: what Zuckerberg said about privacy, pricing, and Llama 4 og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Meta's Muse AI agent: what Zuckerberg said about privacy, pricing, and Llama 4 **[Collections](https://daily.dev/sources/collections)** · 4 min read · 1 upvotes · 0 comments ## Summary Meta launched Muse, a persistent personal AI agent designed for ongoing tasks like meal planning, health tracking, and project management rather than one-off chatbot queries. Mark Zuckerberg discussed the launch with journalist Alex Heath, detailing a privacy architecture where each agent runs in a confidential virtual machine that Meta claims it cannot see into, built with input from Signal founder Moxie Marlinspike, though technical details remain unpublished. Muse runs on a new model called Muse Spark 1.3, with a future model codenamed Watermelon coming from Meta's Prometheus cluster. Muse is free up to about 100 million tokens per week, with paid tiers for additional capacity but no announced pricing. Zuckerberg also called Llama 4 'off the trajectory' Meta needed and criticized labs that keep advanced models closed as 'quite dangerous,' seemingly aimed at Anthropic and OpenAI. ## Content ## What is Muse? Muse is Meta's new personal AI agent, launched this week alongside an interview Mark Zuckerberg gave to journalist Alex Heath on the Sources podcast. Unlike chat-based AI tools, Muse works on ongoing goals - things like meal planning, reviewing gym footage, or auditing your software subscriptions - and connects to email, calendar, and other apps to actually do things rather than just answer questions. Zuckerberg named his own agent "Grippa" after a Roman general, paired with a teddy bear avatar. A colleague named theirs "PIP." The internal culture around personalizing agents with names and avatars seems to be a deliberate design choice rather than a quirk. ## Privacy architecture The most technically interesting part of Muse is how it handles privacy. The agent runs inside confidential virtual machines, meaning Meta itself cannot see what you input or what memories the agent stores. Zuckerberg drew a direct parallel to WhatsApp's end-to-end encryption, framing the confidential VM approach as the same kind of architectural commitment - not just a policy promise. The security setup also includes a secure credential store, one-time virtual credit card numbers for transactions, and a separate "sentinel" agent that monitors actions and flags decisions that need human approval before proceeding. The confidential VM design was built with Signal founder Moxie Marlinspike involved, though the technical details haven't been published yet, so the privacy promise remains unverified for now. ## Pricing and the model underneath Muse is free with roughly 100 million tokens per week included. There's an optional paid subscription for heavier use. The app already has dormant code for an invite system that would give users extra free usage when they redeem invite codes - suggesting a global expansion push may be coming around Meta Connect. The model powering it, Muse Spark 1.3, shipped recently. Zuckerberg admitted that Llama 4 was "off the trajectory" Meta needed. A new model codenamed Watermelon is coming from Meta's Prometheus cluster. Early observers noted that Spark 1.3 is better-behaved than expected given the pace of development. On the performance side: most interactions are reportedly under 150ms, which is fast enough to feel responsive rather than agentic-slow. ## The broader argument Zuckerberg is making Zuckerberg used the launch to make a pointed argument about AI development. He said labs training advanced models without releasing them is "quite dangerous," a comment clearly aimed at Anthropic and OpenAI without naming them. His framing: concentrated AI capability is itself a risk, and putting powerful AI in people's hands is the antidote. He drew a parallel to social media giving everyone a voice, positioning Muse as the same kind of democratizing move for intelligence. Whether you find that argument convincing probably depends on how you feel about Meta's track record - more on that below. ## How the launch actually went Muse hit No. 2 on the US App Store with over 83,000 iOS downloads on launch day. That sounds good until you compare it to Meta's own history: Threads pulled 4.3 million downloads on day one, and even Meta AI managed 108,000. On Android, Muse ranked No. 338 in Productivity on Google Play. The launch also came shortly after Meta's $18 billion multistate settlement over social media harms to children, which isn't a great backdrop for a product asking users to hand over access to their email, calendar, and financial accounts. The main competitor in the consumer agent space right now is Instinct, valued at $2.5 billion with $350 million raised. Instinct is moving fast - rolling out email addresses, Stripe and 1Password integrations, location sharing, and agent-to-agent communication between users' contacts. It has its own privacy and security concerns, but it's shipping. ## What Zuckerberg says he uses it for In the interview, Zuckerberg mentioned using Muse for parenting activity planning, MMA training feedback, and health tracking. He also described a future vision of "business in a box" setups where agents handle most of the operational work for small businesses, and personalized health treatments informed by continuous agent-gathered data. Those use cases are genuinely interesting. Whether Muse gets there depends on whether people trust Meta enough to let an agent this deep into their lives - which, given the settlement news and the still-unpublished privacy architecture details, is an open question. ## Questions this post answers ### What model is currently powering Meta's Muse AI agent? Muse currently runs on a model called Muse Spark 1.3. A newer model codenamed Watermelon, developed from Meta's Prometheus cluster, is coming next and is intended to correct the direction Zuckerberg said Llama 4 took, which he described as 'off the trajectory' Meta needed. _Track how Meta's agent models evolve past Llama 4 by following AI agent coverage on daily.dev._ ### How much does Meta's Muse AI agent cost to use? Muse is free for roughly 100 million tokens per week per user. A paid tier exists for additional capacity beyond that free allotment, but Meta has not announced exact pricing for the paid plan. _Compare AI agent pricing models like this one as they roll out via daily.dev._ ### How does Meta claim to protect user privacy in the Muse AI agent? Each Muse agent runs inside a confidential virtual machine that Meta says it cannot see into, similar to WhatsApp's end-to-end encryption, storing memories accessible only to that user's agent. The design also includes a secure credential vault, one-time virtual payment cards, and a sentinel oversight agent flagging sensitive actions for human review, built with input from Signal founder Moxie Marlinspike, though technical details remain unpublished. _Weigh unverified privacy claims like this one before trusting an AI agent with personal data, a call daily.dev helps developers make._ ## Community take How the wider developer community reacted, aggregated from 1 discussion and 16 comments across x (as of 2026-09-16). **TL;DR:** The available replies mostly veer off into a tangent about UI latency thresholds (the 150ms/400ms 'Doherty threshold' debate) rather than discussing Muse's privacy, pricing, or Llama 4 angle directly, with a few scattered comments praising Muse's speed compared to other chatbots. **Sentiment:** 40% positive · 50% mixed · 10% skeptical **The case for** - A few replies say Muse feels notably faster than other chatbots like Grok. - One commenter appreciated the low-latency, responsive feel of the product. **The pushback** - One reply notes that basic layout/performance issues persist elsewhere on Facebook despite this speed focus. **By community** - x (mixed): Discussion drifts into a technical tangent about interface latency thresholds, with only light, mostly positive commentary on Muse's speed. **Hottest debate:** Whether the relevant human-perceptible lag threshold is 150ms or the 400ms 'Doherty threshold'. **Highlights** > @jaredpalmer @natfriedman For those not knowing, ~150 ms is how long it takes for humans to identify lag. At Apple we were obsessed with keeping it under that number > — [erfahrung on x · 8 points, 2 comments](https://x.com/erfahrung/status/2100328879894528034) > @erfahrung @jaredpalmer @natfriedman The Doherty threshold is 400ms > — [birkirgudjonson on x](https://x.com/birkirgudjonson/status/2100332148318224474) > @jaredpalmer @natfriedman I’ve noticed how much slower the other chatbots are now that I’ve had a coupe of days interacting with Muse. Great product. > — [calcaneus1326 on x](https://x.com/calcaneus1326/status/2100359408391737688) > @jaredpalmer @natfriedman The speed difference vs instinct/grok bot is significant > — [patrickdonohoe on x](https://x.com/patrickdonohoe/status/2100360761004413121) **Source threads** - [x](https://x.com/jaredpalmer/status/2100290668237140479) · 0 points · 16 comments ## Similar posts on daily.dev - [Meta debuts its Muse AI agent. Will consumers trust it?](https://daily.dev/posts/meta-debuts-its-muse-ai-agent-will-consumers-trust-it--nuqtze2hd) · TechCrunch · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#privacy](https://daily.dev/tags/privacy), [#llama](https://daily.dev/tags/llama) [View this post on daily.dev](https://daily.dev/posts/meta-s-muse-ai-agent-what-zuckerberg-said-about-privacy-pricing-and-llama-4-y8zu7g1wa) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Meta's Muse AI agent: what Zuckerberg said about privacy, pricing, and Llama 4","url":"https://daily.dev/posts/meta-s-muse-ai-agent-what-zuckerberg-said-about-privacy-pricing-and-llama-4-y8zu7g1wa","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/meta-s-muse-ai-agent-what-zuckerberg-said-about-privacy-pricing-and-llama-4-y8zu7g1wa"},"datePublished":"2026-09-10T15:38:29.005Z","dateModified":"2026-09-16T23:45:24.723Z","description":"Meta launched Muse, a persistent personal AI agent designed for ongoing tasks like meal planning, health tracking, and project management rather than one-off...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/6e6f05a6279e167e61f26ec25e96a0d0?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/6e6f05a6279e167e61f26ec25e96a0d0?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/meta-s-muse-ai-agent-what-zuckerberg-said-about-privacy-pricing-and-llama-4-y8zu7g1wa","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents,privacy,llama","timeRequired":"PT4M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Meta's Muse AI agent: what Zuckerberg said about privacy, pricing, and Llama 4"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/meta-s-muse-ai-agent-what-zuckerberg-said-about-privacy-pricing-and-llama-4-y8zu7g1wa#faq","mainEntity":[{"@type":"Question","name":"What model is currently powering Meta's Muse AI agent?","acceptedAnswer":{"@type":"Answer","text":"Muse currently runs on a model called Muse Spark 1.3. A newer model codenamed Watermelon, developed from Meta's Prometheus cluster, is coming next and is intended to correct the direction Zuckerberg said Llama 4 took, which he described as 'off the trajectory' Meta needed. Track how Meta's agent models evolve past Llama 4 by following AI agent coverage on daily.dev."}},{"@type":"Question","name":"How much does Meta's Muse AI agent cost to use?","acceptedAnswer":{"@type":"Answer","text":"Muse is free for roughly 100 million tokens per week per user. A paid tier exists for additional capacity beyond that free allotment, but Meta has not announced exact pricing for the paid plan. Compare AI agent pricing models like this one as they roll out via daily.dev."}},{"@type":"Question","name":"How does Meta claim to protect user privacy in the Muse AI agent?","acceptedAnswer":{"@type":"Answer","text":"Each Muse agent runs inside a confidential virtual machine that Meta says it cannot see into, similar to WhatsApp's end-to-end encryption, storing memories accessible only to that user's agent. The design also includes a secure credential vault, one-time virtual payment cards, and a sentinel oversight agent flagging sensitive actions for human review, built with input from Signal founder Moxie Marlinspike, though technical details remain unpublished. Weigh unverified privacy claims like this one before trusting an AI agent with personal data, a call daily.dev helps developers make."}}]} ```