Connect external AI agents like Claude, Instinct, and Muse to your Fabric workspace. Fabric treats them like non-human team members with granular access controls, roles, and permissions. They get exactly the access you grant, nothing more.

External agents, managed like team members.

Invite any agent.
Connect external AI agents to your Fabric workspace the same way you would invite a colleague. Claude, Instinct, Muse, and any other external agent can be added and given a role with defined permissions. Fabric becomes the control plane between the agent and your data.

Non-human team members.
External agents are treated the same way as human collaborators. They have roles, access levels, and defined permissions. They can see the content you grant them access to and nothing else. The same collaboration and permission model that governs your human team members governs your agents.

Separate from Fabric's own agents.
Fabric has its own native background agents that you configure, schedule, and run within your workspace. Agent Vault is specifically about external agents, the third-party AI tools you invite in, and controlling what they can access and do once they arrive.

Granular access to your workspace.

Folder and project-level permissions.
Grant external agents access to specific folders, spaces, or projects within your workspace. An external agent helping with a client project gets access to that project folder. It does not see your other clients, your financials, or your internal strategy docs.

Time-limited access.
Approve access for a single task, for a day, for a week, or indefinitely. When the access window expires, the external agent can no longer read or write to that content. You do not need to remember to revoke access manually.

Capability controls.
Beyond which content an agent can see, you control what it can do. Reading files, creating documents, moving content, deleting items, running searches. Each capability is a separate permission you grant or withhold.

The trust problem with external agents.

External AI agents are becoming more capable and more autonomous. Claude can research, write, and execute multi-step workflows. Other agents can analyze data, manage projects, and interact with your tools. The value of connecting these agents to your workspace is enormous. But so is the risk of giving them uncontrolled access.

Today, connecting an external agent to your data usually means giving it broad access to everything. There is no way to say "you can read this folder but not that one" or "you can create documents but not delete them." The agent either has access or it does not. This all-or-nothing model stops teams from using external agents with their most valuable content, which is precisely where agents would provide the most value.

Fabric solves this by acting as the control plane between external agents and your data. When you invite an external agent to your workspace, Fabric treats it like a new team member joining the organisation. You set its role, define what it can access, and specify what capabilities it has. The agent operates within those boundaries. Your workspace, your notes, your self-writing docs, your client materials, your research, all of it remains under your control even when external agents are working with it.

Permission requests, not assumptions.

When an external agent needs content it does not currently have permission to access, it sends a permission request through Fabric. You receive a notification describing what the agent needs and why. You choose how to respond: approve for one time, approve for a limited period, approve indefinitely, or deny. The agent proceeds once you respond and waits if you do not.

This means you do not have to anticipate everything an external agent might need when you first invite it. Start with narrow permissions. Let the agent request additional access as it encounters content it needs. You make the access decisions in real time, informed by what the agent is actually trying to do. For particularly sensitive content like client documents, financial data, or legal materials, you can require that access is always requested and never pre-granted, ensuring a human approval step for every interaction.

Full audit trail.

Every action an external agent takes in your workspace is logged. What content it read, what it created, what it modified, and when. The audit log is available for review at any time so you can verify exactly what external agents have been doing with your data. The log includes the context: which task prompted the access, what permission grant the agent was operating under, and what it produced as a result.

For teams, the audit trail is essential. Administrators can review external agent activity across the entire workspace, see which agents are accessing which content, and identify anomalies. Notifications can be configured for unusual access patterns so problems surface early. This gives organisations the confidence to invite external agents into their workspace at scale without losing visibility into what those agents are doing with company data.

Built for teams, not just individuals.

Agent Vault is especially powerful for organisations. A startup can invite Claude to help with engineering documentation and limit its access to the engineering workspace. An agency can invite a specialized research agent and scope its access to a single client project. A law firm can invite an analysis agent and restrict it to specific case files with time-limited access that expires when the analysis is complete.

Team administrators set organisation-wide policies about which external agents can be invited, what roles they can be given, and what maximum permissions they can hold. Individual team members work within those policies. The company maintains governance over external AI access while individuals and teams get the benefit of powerful external agents working alongside them.

Use cases

External agent for a specific project
Invite Claude to help with a research project. Grant it access to the project folder and the relevant research materials. It reads the sources, produces analysis, and creates documents within that folder. It never sees content outside the project. When the project ends, revoke the access.

Specialized agents for specialized work
Different external agents are good at different things. Invite a coding agent for engineering work, a research agent for literature review, and a writing agent for content planning. Each gets access to only the content relevant to its task.

Compliance and governance
Maintain a complete record of every external agent's access to your workspace. Demonstrate that external AI tools interact with company data under controlled, auditable conditions. Meet data governance requirements while still benefiting from external AI capabilities.

Gradual trust building
Invite an external agent with narrow, time-limited access. Review what it does. Expand permissions as you build confidence. You do not need to make an all-or-nothing trust decision on day one.

Perfect for

Teams
Invite external agents and manage them like non-human team members. Administrators set policies, individuals deploy agents within those boundaries, and the organisation maintains full visibility. Learn more about Fabric for teams.

Agencies and consultancies
Client confidentiality demands that external agents working on one client's materials never access another client's content. Agent Vault enforces this with project-level access controls and complete audit trails. Learn more about Fabric for agencies.

Law firms
External agents can assist with analysis and research, but access to case files must be controlled, time-limited, and fully auditable. Agent Vault provides the access governance that legal work requires. Learn more about Fabric for law firms.

Startups
Use the best external AI agents available without worrying about giving them free access to everything. Start narrow, expand as you build trust, and maintain full control as your team and your agent usage grow. Learn more about Fabric for startups.

Works seamlessly with other features.

Background agents
Fabric's own native background agents operate within your workspace with their own permission model. Agent Vault is the equivalent control layer for external agents that you invite into your workspace.

Private and secure
Agent Vault extends Fabric's privacy and security model to external AI agents. The same principles of controlled access, auditability, and revocability that protect your content from unauthorized human access apply to every external agent.

Collaboration
External agents slot into Fabric's collaboration model as non-human team members. Roles, permissions, and access controls work the same way whether the team member is human or AI.

Agent Passes
Agent Vault controls what external agents can access within your Fabric workspace. Agent Passes controls what those agents can do in your connected external apps. Together, they provide complete governance over external agent behavior across your entire ecosystem.

FAQ

What is an "external agent" in this context?
Any AI agent that is not Fabric's own native background agent. This includes agents like Claude, Instinct, Muse, and any other third-party AI tool that can connect to and interact with your workspace.

How do I invite an external agent to my workspace?
Add the agent through Fabric's setup process, similar to inviting a new team member. You assign it a role and define its permissions. The agent connects to your workspace through Fabric's control layer and operates within the boundaries you set.

How is this different from Fabric's own background agents?
Fabric's native background agents are built into Fabric. You create them, write their prompts, and schedule them. Agent Vault is about external agents built by other companies that you invite into your workspace. Fabric treats them as non-human team members and gives you the same level of access control you have over human collaborators.

Can I have multiple external agents with different permissions?
Yes. Each external agent is a separate non-human team member with its own role and permissions. A research agent might have read access to your library. A writing agent might have read and write access to a specific project. A coding agent might only see your engineering spaces.

What happens when an external agent requests access to something?
You receive a notification describing what the agent needs and why. You approve for one time, a limited period, or indefinitely. You can also deny. The agent waits for your response before proceeding.

Can I revoke access after granting it?
Yes. Any permission can be revoked instantly. Time-limited grants also expire automatically. The external agent immediately loses the revoked capability.

Can I see everything an external agent has done?
Yes. The audit log records every action: what content was read, created, modified, or deleted, and when. You can review the log at any time and filter by agent, content area, time period, or action type.

Can team administrators control which external agents are allowed?
Yes. On Team plans, administrators set organisation-wide policies about which external agents can be invited, what roles they can hold, and what maximum permissions they can be granted.

Can I get alerts about unusual behavior from an external agent?
Yes. Configure notifications for anomalous patterns, such as an agent accessing significantly more content than usual or attempting to access content areas outside its normal scope.

Is Agent Vault available on all plans?
Agent Vault is available on Pro and Team plans. Advanced team governance features are available on Team plans. See individual pricing or team pricing for details.