QX Labs vs Meta Muse: Personal AI Agent or Business Platform?

The short answer: Meta Muse is a personal AI agent for one person. It books, buys and emails on your behalf from a private cloud computer, you talk to it in the Muse app or on WhatsApp, and it is only available in the United States. QX Labs is an AI agent and automation platform for teams. You build as many agents as you need, decide tool by tool what each one may touch, reach the same agent from Slack, Microsoft Teams, WhatsApp, email or the web without it forgetting anything, share it with colleagues, and put it to work across 1,000+ apps, Grids and Flows. On WhatsApp specifically, QX can also put an agent on your own business number to answer customers. Muse only answers you.

Muse launched on 8 September 2026 and has had a busy fortnight: Muse for Mac, a connector platform for developers, a wave of retail partners at Meta Connect on 23 September, a block from Amazon, and a privacy story that went round the industry. We explained the product itself in What is Meta Muse?. This post is the comparison: where the two overlap, where they differ, and which one fits what you are trying to do.

Key takeaways

• Muse gives you one agent, yours alone, that acts on your personal accounts. QX gives a team a roster of agents, each with its own brief, tools, knowledge and @name.
• Control works at different grain. Muse grants access per connector (read, or read and send) and a supervisor approves each network request. QX scopes an agent down to individual tools within an app, one designated account, and specific folders in a knowledge vault, with every choice visible in the editor.
• Muse lives in its app, muse.ai, WhatsApp and now the Mac. QX agents answer in Slack, Teams, WhatsApp, email, the web app and via API, with one history and one memory behind all of them.
• Muse has no sharing and no team. QX agents can be private, shared with your whole workspace, shared with named people, or published by link.
• Muse connects to a couple of dozen consumer services, from Gmail to Walmart. QX connects to 1,000+ business apps and adds Grids for row-scale work, Flows for scheduled pipelines, and Knowledge Vaults for cited answers.
• On WhatsApp, Muse is a personal assistant you message. A QX external agent sits on your own WhatsApp number and replies to customers 24/7 from your price list and calendar.

What is Meta Muse?

Muse is Meta's personal AI agent. Each user gets a dedicated virtual machine in Meta's cloud with its own browser, so the agent can open websites, fill in forms, negotiate a bill or check out with a single-use card from Stripe Link. A separate supervisor process, Sentinel, holds your credentials and approves or blocks every action the agent proposes, asking you first for sensitive ones like sending an email or paying for something. It is free for most use, with Power at $20 a month and Maximum at $100, and it is limited to adults in the US. It runs on Meta's own Muse Spark model and nothing else.

Since launch the surface has grown quickly. Meta opened a connector platform so developers can submit integrations, and said it received more than 1,500 applications in under a week. Muse for Mac arrived with computer use, meaning the agent can drive any app on your Mac with your permission. At Meta Connect on 23 September, Meta announced retail connectors for Walmart, Best Buy, Gap, Sephora, Wayfair and others, plus Expedia, Instacart, Notion, Granola, GitHub and Box, an email address of Muse's own, a glasses integration, and a keychain device called Muse Charm shipping before the holidays. Zuckerberg was clear about the business model: Meta "will profit by taking a small fee from transactions."

Two other things happened that matter for this comparison. Amazon blocked Muse from shopping on its site on 21 September, saying the agent does not identify itself while browsing and appears to capture and store customer credentials. And an Inc. columnist, Jason Aten, reported that Muse for Mac had synced his entire iMessage database after he declined Messages access during setup, then told him it was only seeing "the incoming notification stream." Meta's David Singleton called the explanation "on us" and said the feature was opt-in. We come back to that below, because it is the clearest illustration of what "control over what an agent can do" means in practice.

What is QX Labs?

QX Labs is an AI agent and automation platform built for teams. The premise is that your next hire is an AI agent: a chatbot answers and leaves you to do the work, rigid automation only follows fixed rules, and QX understands context, uses your real tools and finishes the job.

It is made of a few pieces that fit together:

• Agents are AI co-workers you brief in plain English. Each one has its own model (OpenAI, Anthropic, Google and others, with your own keys if you like), instructions, capabilities, connected apps and knowledge.
• Grids run an agent, an integration or a formula across hundreds or thousands of rows in parallel.
• Flows are triggered or scheduled workflows that mix fixed steps with agent judgement, with approval gates where actions are irreversible.
• Knowledge Vaults index and continuously sync your documents so agents answer from your material, with citations.
• 1,000+ integrations give agents reach across CRMs, inboxes, calendars, documents, finance tools and the long tail of research and enrichment APIs.
• An API to run agents, flows and grids from your own software.

A free plan includes all of it. QX Labs is a UK company, ISO 27001:2022 certified, GDPR aligned, and not limited to one country.

QX Labs vs Meta Muse at a glance

	Meta Muse	QX Labs
Built for	One person's personal life	Teams and businesses
Number of agents	One Muse per user	As many as you need, each with its own role and @name
What it can act on	Your personal email, calendar, shopping and finance accounts	Your company's tools, documents and data
Control over access	Per connector (read, or read and send); Sentinel approves each request	Per tool within an app, per account, per vault folder; capabilities toggled one by one; per-conversation overrides
Where you talk to it	Muse app, muse.ai, WhatsApp, Mac; glasses and Charm coming	Slack, Teams, WhatsApp, email, web app, API
Memory across channels	One personal agent, one memory	One agent, one history and workspace behind every channel
Sharing	None	Private, whole workspace, named people, or public link
Integrations	Roughly two dozen consumer connectors plus a developer platform	1,000+ business apps
Work at scale	One task at a time	Grids across thousands of rows; scheduled Flows
Customer-facing WhatsApp	No	Yes, on your own number, external agents
Model	Muse Spark only	OpenAI, Anthropic, Google and more, chosen per agent
Availability	US, 18+	Not region-locked
Pricing	Free, $20, $100 a month per person	Free plan with every feature; usage credits shared by the workspace

Muse details as described by Meta and press coverage between 8 and 23 September 2026. It is changing weekly, so check Meta's announcement for the current state.

How much control do you have over what the agent can do?

This is the question the Muse launch has forced into the open, and the two products answer it at different levels.

Muse's model is permission per connector plus a supervisor. When you link Gmail you choose read-only or read and send. Sentinel then sits between the agent and the internet, checking every request against policy and pausing for your approval before a purchase or an outgoing email. The credentials live in a vault the model cannot read. It is a serious design for a consumer product, and we said so in the explainer.

The Mac release shows the limit of connector-level permissions. Muse for Mac asks macOS for Full Disk Access so it can organise your downloads and find lost files. Aten declined Messages access in setup and still found the agent had synced more than 187,000 rows of his iMessage database. Whatever the cause, the access was coarse enough that "organise my files" and "read every text I have ever sent" were the same permission. Amazon's complaint runs the other way: from the merchant's side, Muse arrives as a browser that does not say it is an agent and appears to hold the customer's credentials.

QX approaches control from the other end, because a work agent that can reach Salesforce and the finance inbox needs finer settings than "connected" or "not connected."

• Tool-level scopes. An agent connected to Gmail can be given the read tools and not the send tool. The docs put it plainly: only the tools you see in its editor exist for it. A disabled tool is not merely discouraged, it is absent, so no amount of persuasion in the chat brings it back.
• One designated account. When several people have connected the same app, the agent acts as one chosen account, so a shared agent never runs against a colleague's private connection.
• Knowledge scoped to folders or files. A vault can be attached whole or narrowed to the folders the agent should see.
• Capabilities toggled individually. Web search, chat memory, scheduled runs and workspace access are separate switches, per agent or per conversation.
• Approval gates in Flows. For anything irreversible, a Flow routes the draft to a person in Slack or email before the send step, and branches on rules like "discount over 15% needs a VP."
• Credentials the model never sees. Integration credentials are encrypted at the application layer and revocable in a click.
• A log of every run. Each step's inputs and outputs are inspectable afterwards, which is what an auditor or a manager asking "why did it email that customer" actually needs.

For customer-facing agents, QX goes further still. An external agent on WhatsApp gets no general app access at all. It gets Actions: one tool, on one account, with the arguments fixed in advance and only the customer-shaped fields (a name, a party size, a preferred time) filled in by the model. "Add this booking to the Bookings sheet" is possible. "Do anything else in Google Sheets" is not, by construction. The customer has no authority over the agent's instructions, and rate limits cap what a stranger can make it spend. The Trust Center documents the certifications behind all of this.

Where can you talk to it, and does it remember?

Muse meets you in its app, at muse.ai, on WhatsApp, and now on the Mac, with AI glasses and the Charm keychain to follow. The WhatsApp version is a plain chat; the app's feed and ideas tools are not there. It is one agent with one memory, and it remembers what matters to you across conversations, which is what you want from a personal assistant.

QX starts from where a team already talks. Install the QX app in Slack or Teams and @mention an agent in a channel, message it on WhatsApp, email it at its own alias, or open the web app. The point is that it is the same agent behind every door. Every thread from every channel lands in one chat history, an agent can search past conversations regardless of which channel they happened on, and the persistent workspace keeps the files it built and what your team has taught it between sessions. Ask an agent in Slack on Thursday about the numbers it pulled for you on WhatsApp on Monday and it can find them. On WhatsApp, typing @AgentName switches between your agents mid-chat.

One agent, or a team of them?

Muse is singular by design. You have your Muse, it knows you, and it does your errands. There is no way to make a second one with a different job.

In QX the unit is the role. A @Pipeline agent briefed on your sales playbook, wired to HubSpot and Apollo, grounded in your ICP document. A @Helpdesk agent that only answers from the support vault, with citations. A @Diligence agent for the deal team with read access to the data room and nothing else. Each carries its own instructions, model, skills, tools and knowledge, and each gets its own @name in Slack, so summoning the right specialist is like tagging a colleague. The default @QX agent works off the shelf while you build the others.

Can you share the agent?

Not with Muse. It is bound to one person, one VM and one set of personal accounts, and there is no concept of a colleague.

QX agents have four visibility settings: private to you, shared with everyone in the workspace, shared with named people, or published by a public link. That is what turns an agent from a personal tool into a hire. The onboarding agent you built answers every new starter. The research agent the analyst set up runs for the whole fund. And because the account swap keeps each user on their own connected accounts, sharing an agent never means sharing your inbox.

Beyond chat: Grids, Flows, knowledge and 1,000+ apps

Muse does one task at a time, in a browser, for you. Its connectors are consumer services: Gmail, Google Calendar, Outlook, Plaid, OpenTable, Spotify, Peloton, Instagram and the other Meta apps, and now a list of retailers. Anyone can submit a connector to the new platform, and Meta says custom connectors it has not reviewed should be granted access "with caution."

QX is an orchestration platform, which changes what an agent is for.

• Grids take a job an agent can do once and run it down a column of two thousand rows: score every inbound lead, research every company in a market, extract fields from a folder of contracts, with the same rubric on every row and every cell's working visible.
• Flows run unattended on a trigger or a schedule: new lead in Salesforce, enrich via Apollo, score, draft outreach in Gmail, notify the owner in Slack, with a human gate before anything external goes out.
• Knowledge Vaults keep the agent grounded in your policies, price lists and playbooks, continuously synced, with citations back to the source.
• 1,000+ integrations cover the business stack from Salesforce and HubSpot to Notion, Google Drive, Stripe, Xero, GitHub and the research and enrichment APIs most no-code tools skip.

An agent you build can be dropped into a grid column or called as a step in a flow, so the same @Diligence brief that answers a question in Slack can also process a hundred pitch decks overnight.

WhatsApp: a personal assistant, or an agent for your customers?

Both products are on WhatsApp, and this is the difference that trips people up.

With Muse, WhatsApp is a way for you to reach your own agent. You message Muse, it acts on your accounts, it replies to you. Nobody else can message your Muse, and it cannot answer on your behalf to the people who message you.

QX supports the same personal use: registered users message the QX WhatsApp number and get their agents. It also supports the opposite direction. An external agent pairs with a WhatsApp number you already own, the same way you link WhatsApp Web, and replies to anyone who messages that number. A regular or a WhatsApp Business number both work. You describe the business, QX reads your website and builds the agent, you upload the price list, FAQs and opening hours, connect the calendar or sheet it may write to, and set the rules for what escalates to a person. From then on a plumber's number answers "do you cover SW11 and what's the call-out fee" at 11pm, a salon's number books the appointment straight into the calendar, and a complaint gets a "Sam will call you back" with a logged callback rather than an improvised answer. Every thread appears in your QX history with each action the agent took, the agent never messages first, and it ignores group chats. There are setup guides for 50 kinds of business, from dental clinics and restaurants to estate agents and recruitment agencies.

To be fair to Meta, it does sell a customer-facing agent. It is a different product called Meta Business Agent, available globally since June 2026 on WhatsApp Business and Instagram DMs, free to start with paid tiers to follow, and an enterprise platform that connects to systems like Shopify and Zendesk. If you only want a replier inside Meta's ecosystem, look at it. The QX version is one agent among your others, grounded in the same vaults, built by the same builder, logged in the same history, and able to act in the same 1,000+ apps through scoped Actions.

Where Meta Muse is the better pick

Muse is a good product for what it is, and there are plenty of cases where it is the right answer.

It is built for personal errands, and it does them with a real browser and a real card. Selling a car, lowering a bill, booking a tennis lesson, buying the thing you saw on Instagram: that is Muse's home turf, and QX is not trying to be your household assistant. The free tier covers most of that. The security architecture, with a per-user VM, a supervisor outside the model, and credentials the agent cannot read, is more serious than anything else published for a consumer agent, and Meta has put it out for scrutiny with a bug bounty up to $300,000. The hardware story, with glasses and the Charm, has no equivalent anywhere in the business tools market. And if Meta's connector platform grows the way its 1,500 early applications suggest, Muse's reach across consumer services will widen fast.

If you are an individual in the US and you want an agent for your own life, Muse is the one to try.

Which should you pick?

Choose Meta Muse if you want one personal agent for errands and shopping, you are happy to run it on Meta's model in Meta's cloud, you are in the United States, and you do not need it to work for anyone but you.

Choose QX Labs if you are a team, if the work touches company systems, if you want more than one agent and want to share them, if the agent must answer in Slack or Teams as readily as WhatsApp and remember across all of them, if you need work at row scale or on a schedule, or if you want a WhatsApp agent that answers your customers rather than you.

Put plainly: Muse is an agent that works for a person. QX is a platform for building the agents that work for a business.

FAQ

Can I use Meta Muse for my business?

Not really. Muse acts on one person's personal accounts, has no team permissions or sharing, and cannot answer customers who message you. Meta's separate Meta Business Agent handles customer replies on WhatsApp Business and Instagram. For agents that work across company tools, documents and channels, you need a business agent platform such as QX.

Does QX Labs work on WhatsApp like Muse does?

Yes, and in both directions. Team members message the QX WhatsApp number to reach their agents, switching between them with @AgentName. Separately, an external agent can be paired to your own WhatsApp number, regular or Business, and reply to customers from your documents, book into your calendar, and hand off to a person under rules you set.

Does a QX agent remember conversations across Slack, Teams and WhatsApp?

Yes. It is one agent behind every channel. Every thread lands in one chat history, the agent can search past conversations regardless of where they happened, and its persistent workspace keeps files and learned context between sessions. Muse also has one memory, but it belongs to one person and one assistant.

Is Meta Muse safe to give access to my accounts?

Its architecture is stronger than most consumer agents: a private VM, a supervisor that approves each request, and credentials the model cannot read. Two September incidents show the limits. Muse for Mac synced a columnist's full iMessage database after he declined access, and Amazon blocked it for not identifying itself. Grant permissions narrowly and review what is enabled.

How does QX limit what an agent can do?

Per tool, per account and per folder. You enable individual tools inside a connected app, so an agent can read an inbox but not send from it. It acts as one chosen account. Knowledge vaults can be narrowed to specific folders. Customer-facing agents get only pre-bound Actions with fixed arguments. Every run is logged step by step.

How much do Meta Muse and QX Labs cost?

Muse is free for most use, with Power at $20 and Maximum at $100 a month per person, US only. QX has a free plan with every feature, and paid plans use workspace-wide credits rather than per-seat fees, so adding colleagues or agents does not add licences. Current QX plans are on the pricing page.

See it for yourself

If Muse has you thinking about what an agent could do for your business rather than your errands, the quickest way to find out is to watch one run on your own data. Book a demo and we will build an agent live on one of your workflows, or start free and put your first agent in Slack, Teams or on your WhatsApp number this afternoon.

Explore the pieces: Agents, Grids, Flows, Knowledge, WhatsApp agents, and the 1,000+ integrations. For the wider field, see our roundup of the best AI agent platforms and the comparisons with Claude Tag and Grok Bot.