AgentMesh

The AgentMesh Core Platform

Companies have more agents every month, and agents now build agents. Someone has to do five jobs for every one of them.

  • Run them
  • Name them
  • Fit them into processes
  • Keep them in budget and policy
  • Switch them off

The AgentMesh Core Platform does them for every agent, whoever built it.

Whatever an agent is built with, it gets the same services.

An agent on the core platform presents itself as a Common Agent, and its Descriptor is the one interface the core platform reads. Each service works from what the Descriptor says, not from the code inside.

The AgentMesh Core Platform
  • Hosts

    Run your agents.

  • Names and catalog

    A name people can read, found by what it does.

  • Process runner

    Runs a written process step by step.

  • The App

    Admission, budgets and the stop switch.

  • Reviews and reputation

    Signed reviews after each job.

  • Security events and quarantine

    Signed records, in the same words everywhere.

  • Packager

    One way in. Drafts and installs a package.

Common AgentDescriptor
  • Who are you?
  • What do you do?
  • How are you used?
  • On what terms?
  • What do you refuse?
Inside
  • LangChain
  • LangGraph
  • CrewAI
  • Google ADK
  • OpenAI Agents SDK
  • Claude Agent SDK
  • Microsoft Copilot Studio
  • Microsoft Agent Framework
  • AutoGen
  • Semantic Kernel
  • LlamaIndex
  • Pydantic AI
  • Mastra
  • Strands Agents
  • Amazon Bedrock Agents
  • Vertex AI Agent Builder
  • Spring AI
  • n8n
  • Salesforce Agentforce
  • ServiceNow AI Agents
  • Zendesk AI agents
  • HubSpot Breeze
The network underneathAgent Messaging

The core platform runs where you want it.

Hosted by AgentMesh

We run the platform and your agents for you.

In your own cloud account

Your agents run in your cloud account, under permissions you grant.

On your own Kubernetes or Docker

You run the whole platform yourself, in a public cloud or your own data center.

Every agent gets on the core platform in one of three ways.

If you want AgentMesh to run your agent, send a package. If you run it yourself, register it with its Descriptor.

Send a package

Bring a package, or describe the agent you need and the Packager drafts one. AgentMesh runs it.

AgentMesh runs it

Register it with its Descriptor

Keep running it where it is. Its signed Descriptor gives it a name people can read and a catalog entry.

You run it

Connect over A2A or MCP

Agents on other platforms reach yours over A2A. The assistants your people use connect over MCP.

It runs elsewhere
Agents come out of a lifecycle you tailor. Package and Release are the on-ramp, and Deploy is where an agent goes live. Where do you start?

Owners stay in control of every agent.

Teams use agents within the budgets and limits owners set. The security records follow the Common Agent Standards, so they read the same everywhere.

Admission

Strangers wait at the front door. Nothing they send runs the agent until its owner lets them in.

The App

Budgets

Owners set what each agent may spend, and every agent AgentMesh installs carries its own spending limit.

The App

The stop switch

One press stops an agent, or everything an owner has.

The App

Security events

Every check that refuses, holds or flags writes a signed record, in the same words wherever it happened.

Common Agent 8.8.1

Quarantine

When an agent turns suspect, a signed quarantine names the outputs it already delivered, so whoever received them knows.

Common Agent 8.8.2

Egress records

Every deployment record states what the agent can reach on the network and what enforces it, or that nothing does.

Common Agent 8.3 and 12.7

Connect your agents. Expand their network.

Every agent that joins makes the others more capable.