After Meta Muse: Why Personal and Enterprise Agents Need to Know You

Concept illustration: an agent becomes useful through context, connected tools, and clear access boundaries.

Meta introduced Muse in September 2026 as a personal AI agent. People can message it in the Muse app or WhatsApp, ask it to browse the web, fill forms, send email, book travel, create content, and keep working on longer tasks after they close the app. Muse asks for approval before sensitive actions and provides an activity trail.

The attention around Muse shows that people are looking for more than a chatbot that answers questions. They want an agent that understands their goals and habits, can use the services they already rely on, and can carry work through.

That is why products like Muse and Leina matter: an agent can only do useful work when it understands the person or organization and has an authorized way to act.

An agent first needs to know who it is working for

Personal life and business work are spread across many systems. A person may use Instagram, WhatsApp, Facebook, Gmail, a calendar, shopping, and payment services. A company may keep customer records in a CRM, orders and inventory in an ERP, and project knowledge in documents, storage, support systems, and a knowledge base.

An AI that stays inside one chat window only sees what someone copies into it. A useful agent needs to connect those systems within an approved scope, keep the relevant context, and join multiple steps together.

flowchart LR
  accTitle: Personal and enterprise agents: context, connections, and permissions
  accDescr: Persistent context, connected systems, and permission boundaries guide an AI agent toward a verifiable action.
  C("Persistent context\ngoals, habits, projects"):::channel --> A("AI agent"):::agent
  I("Everyday systems\npersonal services and business Apps"):::resource --> A
  P("Permission boundaries\nwho can see and do what"):::gateway --> A
  A --> R("Verifiable action\nresult, sources, next step"):::resource

Without context, the agent relearns everything. Without connections, it can only suggest. Without permissions, every new connection increases the risk of reaching too far.

What Muse adds to the personal-agent model

According to Meta’s official announcement, Muse focuses on four capabilities:

• It can turn a goal into a plan, continue multi-step work, and return for approval when needed.
• It can use a browser and connected services for email, travel, shopping, reminders, and web tasks, with WhatsApp as one of its entry points.
• It can remember useful details, such as turning an Instagram recipe into a grocery list and remembering dietary restrictions.
• It puts controls into the execution path through Muse Secure VM and Sentinel. People can review activity, adjust permissions, disconnect services, and approve sensitive actions.

This moves a personal assistant from conversation toward action. It can open a browser, call a service, wait for a change, and return with the result under the user’s direction.

Personal and enterprise agents have different boundaries

The underlying capabilities are similar, but the management problem changes with the setting.

An individual usually manages their own accounts and data. An organization must split “what the agent knows” and “what the agent can do” across roles. A sales member may need customer follow-up records; a project group may need project documents; an external group may only receive approved progress. A chat boundary cannot replace permissions in the underlying business system.

Leina brings agent capabilities into team work

Leina is an AI teammate that works in team chat. It remembers team habits, project context, and unfinished work, connects email, documents, code, calendars, CRM, support, data, and storage through Apps, and uses Skills to make proven workflows reusable.

This complements Muse’s personal-agent focus. Muse brings a person’s goals into connected services. Leina brings the same core ideas into multi-person collaboration and business systems.

1. From knowing a person to knowing a piece of work

Leina’s Memory can retain team habits, project context, and work in progress. Group chats and direct chats have separate working contexts, while longer and scheduled tasks can continue in the background. For a business, the useful memory is not only someone’s preference; it is how a task is done, where the last run stopped, and what needs checking next.

2. Apps turn suggestions into business actions

Leina connects to business systems through team-authorized Apps. Its current product materials cover communication, documents, code, CRM, support, data, and storage. At the time of publication, the OpenConnector gateway supports more than 1,500 providers and 16,000 actions; actual access still depends on the connector, connected account, and organization settings.

For example, a team can ask Leina to read this week’s meeting notes, summarize decisions and follow-ups, and include source links. After review, it can prepare the content for a CRM or project table. Read first, verify, then write: this keeps agent actions observable.

3. Skills turn one successful run into a team method

Muse can plan around a personal goal. Enterprises also need to preserve recurring methods: how to assemble a weekly report, which fields to check in a customer follow-up, who reviews a launch checklist, and where research should be saved.

Leina Skills store task steps, required Apps, and checks so a team can reuse and improve them. The value is the complete workflow: context, tools, and verification steps together.

4. Different people can receive different answers

This is central to enterprise use. An administrator connects business accounts and configures available actions, then assigns access by member and group. Members do not need the account password or raw token; Leina calls tools within the configured authorization.

flowchart TB
  accTitle: One Leina, separate access scopes
  accDescr: One Leina routes calls through OpenConnector to sales, project, and external-group resources while blocking unauthorized data and actions.
  M("Members and groups"):::channel --> L("One Leina"):::agent
  L --> G("OpenConnector\nauthorization and calls"):::gateway
  G --> S("Sales: customer records"):::resource
  G --> T("Project group: project material"):::resource
  G --> E("External group: approved progress"):::resource
  G -.-> X("Unauthorized data and actions"):::restricted

The same request can produce different results for different people because their authorized scope is different. That is what allows an agent to work inside a real organization.

Which agent should you start with?

Use the ownership of the task as the guide:

• If the goal and data belong to one person, start with a personal agent such as Muse.
• If the goal belongs to a team and the data is spread across CRM, ERP, documents, or a knowledge base, start with a team agent such as Leina.
• If a company needs both, begin with a small read-only task, save the verified method as a Skill, and expand access by responsibility.

The shared starting point is simple: define the goal, connect only the systems required for it, and state the access scope and approval points before execution.

Start with a task you can verify

Try a request like this in Leina:

From the connected project documents and CRM records, summarize this week’s customer progress. Read only the relevant sources, include source links, completed work, risks, and next steps. Return a draft first and do not modify any records.

The request names the sources, time range, output, and operation limit. Check the sources, the member’s access, and the destination. Once the workflow is reliable, save it as a Skill or schedule it.

Muse shows how a personal agent can understand goals and act on a person’s behalf. Leina focuses on what happens when an agent enters team chat, business data, and multi-person collaboration: it keeps context, connects tools, reuses methods, and gives each person the scope they are authorized to use.

Personal and enterprise agents will continue to converge. The design questions remain concrete: who is the agent working for, what can it connect to, and what is it allowed to do?

Sources

1. Meta Newsroom: Introducing Muse: The World’s First Personal AI Agent Built for Everyone (September 8, 2026)
2. AI at Meta: Muse features & capabilities
3. Leina homepage, Permissions and Data Access, Apps