As agent adoption accelerates, team dynamics are changing. Teams are increasingly a mix of humans and AI agents. While mature security practices like Identity and Access Management (IAM), least-privilege, and fine-grained permissions are well-established for human participants, organizations are still in the early stages of defining security best practices for agents. MeshAgent’s Room API, participant tokens, and API scopes bring those proven patterns to agents, enabling safe, scalable adoption.
At MeshAgent, we take a simple stance: treat agents as first-class participants alongside humans. Both humans and agents connect to Rooms (our collaborative environments), need clearly defined permissions, and should follow least-privilege principles. This lets you set agent permissions by asking the same questions you do for people:
The MeshAgent Room API provides various capabilities for building agentic apps, but not every participant needs every capability. For example, the Storage API allows you to read and write files, this is essential for participants who create documents or manage artifacts as part of their workflow. However, granting broad file access is risky. With scoped permission to specific paths, you can lock down access so each agent writes only to allowed paths and can read just the information it needs.
Rooms give everyone a shared place to work while scoped participation ensures each person or agent only touches what they should.
Access control is built into every layer of MeshAgent. When you invite a human or deploy an agent or tool you define exactly what that participant can do and MeshAgent enforces it.
First, Rooms provide isolation. Sensitive work lives in its own room, so your finance analytics agents never need (and never get) access to an HR room.
Next, API-level controls ensure participants only see the parts of the platform they actually need. You choose the capabilities required for that participant (e.g., storage, databases, containers, queues, messaging, secrets) and anything not granted is unavailable.
Finally, resource-level restrictions let you set fine-grained controls for each API. For example, an agent might have database access, but can only read from specific tables and not write to them.
Each participant receives a signed token that carries these scopes across MeshAgent. Requests outside the token’s scope are denied automatically, so humans, agents, tools, and services only touch the rooms and capabilities they’re meant to use. MeshAgent automatically logs all agent actions and tool invocations, creating audit trails for compliance and security while enabling you to debug issues and explain agent behavior.
Apply existing access control practices to a new type of participant
You don't need to invent new security paradigms. The same principles you use for managing human access work for agents too. MeshAgent makes it easy to apply them consistently.
Scale your agent deployments confidently
As you build more agents for different use cases, clear permission boundaries prevent the complexity spiral. Each new agent gets permissions for exactly what it needs, preventing unintended access to other systems.
Experiment safely without production risk
Test new agent capabilities in sandboxed rooms with limited permissions before going live. Start with minimal access and expand only as needed, just like onboarding a new team member.
MeshAgent's permission model gives you the control you need to deploy AI agents confidently. Whether you're automating internal workflows or building AI-powered products, you can ensure that every agent, user, and tool has exactly the access it needs and nothing more.
Ready to see how it works? Check out our documentation on Participant Tokens and API Scopes, or contact us to discuss your security requirements.
About MeshAgent: MeshAgent is an AI agent platform that provides complete application infrastructure through collaborative "Rooms" where humans, agents, and tools work together with enterprise-grade security built in from day one.
The MeshAgent Platform is comprised of a powerful three-part system for building and running intelligent agents.
