Skip to main content
This guide explains the core building blocks that make MeshAgent a powerful platform for building, deploying, and operating AI agents — collaboratively, securely, and at any scale. By the end of this document you will understand how Agents, Tools, and Rooms work together to create seamless AI powered experiences.

​
At a Glance: Agents, Tools, and Rooms

MeshAgent enables people, AI agents, and external capabilities (tools) to meet in a secure Room where they can collaborate in real-time, share context, and work together toward common goals.
  • Rooms: Temporary, secure collaboration spaces that automatically spin up when the first participant joins and are retired when the last one leaves. All context, conversations, and shared resources live within the Room. Rooms are provisioned on demand and can scale up or down as your demand changes.
  • Agents: Typically, AI Agents are language model (LLM) powered entities that can understand natural language, make decisions, and execute goals over multiple iterations. Agents are typically given a persona (aka system prompt or rules) and access to tools to help them understand and perform their job. Agents can be chat or voice forward, or run in the background.
  • Tools: Tools extend a model’s capabilities by allowing the model to take action on your behalf — something an LLM itself cannot do. Tools can be registered in the Room and called by human and agent participants. Agents can be equipped with specific tools, and human participants can invoke registered tools.
Deploying with Services: Agents, tools, and external code are deployed as MeshAgent Services — making them available either across all rooms in a project (project services) or scoped to specific rooms (room services).

​
Rooms: The Secure, Scalable Collaboration Hub

​
What are Rooms?

MeshAgent Rooms are secure, on-demand workspaces where humans, agents, and tools collaborate with shared context in real time. Each Room comes with built-in APIs — database, storage, messaging, queues, secrets, and more — so your agent applications have everything they need in one place. See all Room APIs →

​
How Rooms Work

  • On-demand and ephemeral
    A room is provisioned when the first participant joins and deprovisioned when the last participant leaves. Rooms only exists to facilitate communication and work during a session so you never worry about idle infrastructure.
  • Auto-scaling & service availability
    Whether you’re running one agent or ten thousand, Rooms scale automatically. MeshAgent handles provisioning and service setup behind the scenes (to make sure all deployed project and room services are available when you need them).
  • Telemetry and persistence when you need it
    While Rooms themselves are temporary, observability logs/metrics and anything you write through the Storage/Database APIs survive the session. The MeshAgent admin console logs session activity for debugging, such as which agents participated, and what tools were invoked. The other Room APIs can be used to keep track of transcripts, uploads, or other state you want to persist.
  • Granular access controls
    Participant tokens carry scoped grants, letting you control who can enter a room and which APIs humans, agents, or services may call.

​
Agents: Your AI Teammates

​
What are Agents?

Agents are LLM-powered entities that understand natural language, make decisions, and execute tasks over multiple iterations until complete. They’re given:
  • Rules or system prompts that define their role (e.g., “You are a document writer who creates technical documentation”)
  • Tools to help them complete tasks (e.g., file handling, web search, email tools, MCP Tools)

​
Types of Agents in MeshAgent

  • Participant-based agents: (e.g. ChatBot, VoiceBot) Appear as participants in a Room, support back-and-forth conversation, and maintain conversation threads.
  • Background/Ambient Agents: (e.g. TaskRunner, Worker, Indexer) These agents run in the background and don’t appear as participants in a Room. They execute tasks based on queue triggers or direct invocations and return results. For example, a Task Runner executes a specific task when given a structured input and it returns a structured output. These background agents can be given to conversational agents as tools.
You can also run agents in your agent framework of choice inside a MeshAgent Room.

​
Tools: Extending Agent Capabilities

​
What are Tools?

Tools are specialized functions that let agents take action, something LLMs can’t do on their own. Tools can be called by both agents and human participants in a Room. MeshAgent groups related tools together into a toolkit, which is passed to the agent for use.

​
Built-in toolkits

MeshAgent ships prebuilt toolkits for common tasks:
  • Document Authoring: Create and edit documents in a Room
  • UI Interactions: Interact with the user through the UI (e.g., create and generate forms to gather additional information from users)
  • Media Generation: Create images, videos, and presentations
  • Document Processing: Convert documents PDF & Office documents into formats LLMs can use

​
Custom tools and toolkits

Build your own tools to extend agent capabilities:
  • Custom Toolkits: Write your own logic. Tools require an initialization and execution function and should return a specific data type (e.g. TextResponse, JsonResponse).
  • Model Context Protocol (MCP) Support: Connect any MCP server to an agent by passing it as a toolkit or creating a MeshAgent service.
  • OpenAI Connectors: Integrate third-party services like Gmail, Microsoft Teams, and Dropbox through OpenAI Connectors.

​
How Tools Work

  1. Tools are packaged into toolkits (collections of related tools)
  2. Toolkits are passed to an agent when the agent is created or on demand
  3. The agent decides when to invoke tools based on a given task
  4. Tool logic executes and results stream back to the agent
  5. The agent incorporates tool results and continues working until the task is complete and the final result is returned

​
Services: How to deploy and reuse Agents, Tools, and External Connections

Services are how you deploy and run agents, tools, or any code in MeshAgent. Once deployed, services are automatically available where you need them.

​
Types of Services

Project Services
  • Available to all Rooms in a project
  • Use for agents/tools that should be broadly accessible
  • Example: A transcription agent that should be available in every meeting room
Room Services
  • Scoped to specific Rooms
  • Use for specialized agents/tools for particular use cases
  • Example: A legal document analyzer only used in specific contract review rooms
MeshAgent Native vs External
  • Native services: Agents and tools you build with MeshAgent
  • External services: Third-party integrations like MCP Servers or OpenAI Connectors that you connect to rooms

​
Deployment Workflow

1. Build and test locally
Develop your agent or tool on your machine, iterate and test it through MeshAgent Studio. 2. Package as a service
Define your service in a configuration file that specifies how MeshAgent should run the service and what permissions it has. 3. Deploy to MeshAgent
Deploy your service to a project or a specific room. MeshAgent will ensure the service is available the next time you open a room where the service is deployed. 4. Use and iterate
Your service appears in rooms and can be invoked by agents or users. Update and redeploy anytime using the MeshAgent CLI.

​
Next Steps