Meta Muse vs Poke (2026)

Connectors reach email, calendars, the car and the home, Instagram, Facebook and Meta ads, each built with the service provider's API and paired with a skill that tells Muse how to use it well. Muse for Small Business adds Asana, Box, Canva, Dropbox, Figma, Granola, HighLevel, QuickBooks, Klaviyo, Lovable, Notion, Shopify, Slack, Stripe and Zoom. For anything else, Muse can write a custom connector of its own, and it calls tools through command line programs and skills it can build in its VM.

Sourceresearch.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-museread 2026-10-12

In Gmail and Outlook the assistant searches, drafts and sends mail, replies in thread with attachments, forwards, files and labels, and it creates and edits calendar events and syncs contacts.

Recipes add read and write access to Notion pages and databases, Linear and Asana projects, Todoist tasks, GitHub repositories, Vercel and Netlify deployments, Supabase, Ramp spend data, Webflow and Oura, authorized by OAuth or an API key during setup.

Any other service joins as an MCP server added by URL, with an API key or OAuth credentials, or tunneled from a local machine, and Poke discovers and calls the tools it exposes.

Sourcepoke.com/docs/managing-integrationsread 2026-10-12

Muse breaks a goal into an actionable plan and runs several long tasks at once, launching swarms of subagents and handling concurrent subagents and cron jobs inside its VM. The model behind it was trained for long trajectories and multi agent coordination. A Goals tab tracks every goal it is pursuing and the plan to reach it, and a user can interrupt or pile on new tasks without waiting for the last one to finish.

Sourceresearch.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-museread 2026-10-12

A single assistant carries each request across email, calendar and connected tools, so one text can ask it to send onboarding email, schedule a kickoff, create a Notion project and set check in reminders. Recipes package the integrations, an automation and onboarding questions a job needs, and automations keep running in the background on Pro and Ultra. Multistep work rests on the assistant's own planning inside the conversation, with no workflow designer or team of agents to lay out.

Sourcepoke.com/docs/creating-recipesread 2026-10-12

Once a goal or task is set, Muse keeps working on a schedule and in response to relevant events, takes the next step and follows up while the app is closed. It messages first when background work turns up something new or needing a decision, holding proactive messages to a high bar by default, and the user can switch that off or dial it up or down. Requests come in through the iOS and Android apps and the web, all connecting straight to the user's own VM.

Sourceintroducing.muse.airead 2026-10-12

Poke is built to message first. Real time automations run in the background on paid plans, reminders and scheduled tasks fire on their own time, and the assistant watches connected services and its memory to step in, for example adding the action items from a Granola meeting to the calendar. Users reach it by text or voice message in Apple Messages, WhatsApp, Telegram and RCS, and real time sync keeps Gmail and Outlook inboxes current.

Sourcepoke.comread 2026-10-12

Every file a user puts into Muse and everything it generates is stored on the user's own VM, the system of record for the account, and connectors bring in mail, calendars and business tools as tasks need them. For a business, connecting Facebook, Instagram and Meta ads means Muse starts out understanding the business and picks up its voice. Answers come from those files and live connections, with no separately managed knowledge base for a team to curate.

Sourceresearch.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-museread 2026-10-12

Answers draw on the user's own mail, calendar, contacts and connected apps, queried as each request runs, with web search for everything else, including maps and directions. Notion pages and databases, GitHub repositories and DeepWiki documentation become further sources once their recipes are added, and Poke can read a photo of a spreadsheet and turn it into a table or chart. The assistant keeps no searchable store of the user's files of its own, so each answer is assembled from those live sources.

Sourcepoke.com/docs/managing-integrationsread 2026-10-12

Muse's memory persists across conversations and lives in files on the user's own VM, where the user can inspect, edit and download it alongside everything else Muse keeps. Durable application state sits in a separate database on the same machine. How long memory is kept, and whether a single memory can be removed without wiping the rest, is left unsaid.

Sourceintroducing.muse.airead 2026-10-12

Poke keeps memories about the person it works for and combines them with connected services to act at the right moment, while the human operators behind Poke Human see only the details of their task, never the chat history, memories or integrations. Deleting the account clears data within about 24 hours, with backups and legal holds taking longer. What a memory holds, how long it lasts and how a single memory is reviewed or removed are left unsaid.

Sourcepoke.com/faqread 2026-10-12

Every connector action passes through Sentinel, which checks it against the user's connector policy and allows it, denies it or stops for approval. Approval cards appear in the app's own interface, outside the conversation, with clear accept and reject actions, and grants are scoped to one connector and use, lasting once, for a session, for a task, for a set time or permanently. By default Muse browses freely and stops for anything hard to undo, such as writing an email or making a purchase, and the user can change that balance.

Sourceresearch.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-museread 2026-10-12

The assistant acts on what the user texts it, drafting and sending email, booking calendar events and running automations in the background. Control comes through the conversation itself, by disconnecting an integration or by asking Poke to trim busy automations, and there is no approval setting or rule that holds a kind of action for sign off. Poke Human brings in a person for hard errands, a service that does the task instead of checking the agent's work.

Sourcepoke.com/docsread 2026-10-12

Each user's VM splits the agent's runtime cell, mapped to an unprivileged host user, from security services that run outside it.

Privilege separated workers hold connector credentials, an authentication service mints surrogate tokens so the agent never sees real ones, and Sentinel is the sole permission authority for connector actions and network egress, with least privilege enforced between processes.

Agentic red teaming and a public bug bounty back the design, paying up to $300,000 and up to $130,000 for a prompt injection that affects one user. Muse carries no SOC 2 report or other attestation, and a Confidential VM that would keep Meta itself out of user data is still with testers.

Sourceresearch.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-museread 2026-10-12

Accounts are personal, tied to one phone number and opened through a Google or Microsoft account, with no team workspace, roles or single sign on. Interaction applies reasonable physical, technical and organizational safeguards, follows Google's Limited Use rules for Workspace data, and keeps data from users on Maximum Privacy out of model training. Poke offers buyers no SOC 2 report, ISO certification or trust center.

Sourcepoke.com/privacyread 2026-10-12

A snippet under the Muse avatar shows what it is working on, and tapping the avatar opens its full activity log alongside every permission the user has approved. The Goals tab lists everything Muse is tracking and its plan to get there, and each approval card names the exact action and its purpose before it runs. The log and the permission history live inside the app, next to the settings that control them.

Sourceintroducing.muse.airead 2026-10-12

Requests and results live in the message thread, where Poke reports back on what it did, and instructions sent through the API land in the same conversation, so the thread doubles as the record. A subscription page tracks usage against the plan's allowance and the next reset, and a status page covers service health. Users get no separate activity log, run history for automations or audit trail of actions taken in connected apps.

Sourcepoke.com/docs/usage-and-resetsread 2026-10-12

Every user gets a dedicated, isolated VM in Meta's cloud, which holds their files, memory and credentials, and limited data leaves it for inference and telemetry. A user cannot choose where that VM runs, and there is no private or self hosted deployment.

Sourceresearch.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-museread 2026-10-12

Poke runs only as Interaction's hosted service, reached through messaging apps and the web, and sign ups are open in every region worldwide. Buyers get no hosting location, region choice, private deployment or self hosted option.

Sourcepoke.com/docsread 2026-10-12

Each connector ships with a skill Meta wrote on how to get the most out of it, and Muse for Small Business adds a named collection of skills and connectors for owners, covering content and ads, customer replies, cash flow and inventory. An Ideas tab and onboarding tips suggest what Muse could take on based on the user's goals. All of this shapes how the single agent works, and there are no separate prebuilt agents or task templates to pick from a catalog.

Sourcemuse.ai/businessread 2026-10-12

Recipes are ready made setups a user installs in one tap, each bundling integrations, an automation and onboarding questions. Examples include a morning summary of assigned Linear issues, a TripIt import of travel plans, Oura sleep and readiness insights, Notion page organization and Gmail inbox management. The directory groups first party and approved community recipes by category, from calendar, email and finance to health, travel, students and developer tools, and creators publish their own from Kitchen after review.

Sourcepoke.com/recipesread 2026-10-12

Meta trained the model at Muse's core itself, tuned for tool calling through command line programs and skills, long context and multi agent coordination, and a separate set of Meta safety models inspects inference. Conversations and tool trajectories are sanitized and used to train new checkpoints by default. A user cannot choose another model.

Sourceresearch.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-museread 2026-10-12

Plans differ by model class. Pro brings frontier models for complex tasks, Ultra runs frontier models on every action, and the free plan falls back to a lighter mode once its allowance is used. Neither the models nor their makers are named, and a user cannot pick one. Cognition has said its own models and infrastructure will power Poke over time.

Sourcepoke.com/pricingread 2026-10-12

Muse is reached through its own iOS, Android and web clients, which connect straight to the user's VM. Muse can extend itself by writing connectors and skills, but outside software has no API or MCP server through which to call it.

Sourceresearch.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-museread 2026-10-12

The Poke API is one endpoint. A POST with a bearer key from Kitchen delivers a message to the user's own assistant, which carries it out like a text, and the examples cover a browser extension that sends a selected page, a GitHub Action that reports a failed deploy and a Typeform or CRM webhook that hands over a new lead. The poke npm package logs in, adds MCP servers and tunnels local ones, and Kitchen publishes recipes. There is no API for reading conversations or managing integrations, recipes or automations.

Sourcepoke.com/docs/apiread 2026-10-12

Meta hardened Muse through internal use, agentic red teaming and a bug bounty, and Muse judges whether a finished background result merits a notification before it pings the user. Both work on the vendor's product. A user has no way to test a goal, check the agent's output against cases or compare changes.

Sourceresearch.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-museread 2026-10-12

A recipe or automation goes to work once installed, and users have no way to run it against sample input, check its output against expectations or compare versions first. When automations fire so often that a user keeps reaching the plan's limit, the fix is to have Poke review and trim the busiest ones, which manages cost and speed, not quality.

Sourcepoke.com/docs/usage-and-resetsread 2026-10-12

Each user's Muse lives on a dedicated Linux VM with a browser, storage, CPU and memory, where it browses the web, fills forms, shops and books, and compiles code it writes. Its web activity appears as the user's own, and every network request it makes passes through Sentinel. Browsing is one of the headline capabilities, with critical steps such as a purchase held for approval.

Sourceresearch.meta.ai/blog/security-and-safety-for-ai-agents-our-approach-with-museread 2026-10-12

Poke works through integrations, web search and messaging, and it can build and publish a website, but it does not drive a browser, a desktop or a phone's apps. Errands that need a phone call, an order or a booking go to a human operator through Poke Human on the Ultra plan.

Sourcepoke.com/faqread 2026-10-12