Skip to main content

AgentMesh Overview

AgentMesh is a control plane for multi-agent AI orchestration. It runs governed BPMN workflows where each step ("service task") dispatches a governed agent run — a Claude Code (or other CLI) session executing with a scoped set of permissions, skills, plugins, and hooks — instead of a person doing the work by hand.

Think of it as three layers stacked on top of each other:

  1. Orchestration — Camunda 8 runs the BPMN process. Each service task in the diagram is one governed step.
  2. Governance — an Agent Profile decides who runs that step and what they're allowed to do: which model, which permissions, which skills/plugins/hooks, which cost ceiling.
  3. Execution — a Worker (a NodeAgent process you run yourself, next to your infrastructure) receives the run, checks out the repository, executes the CLI, and reports back an auditable result.

Why this exists​

Handing an AI agent a repository and a vague instruction doesn't scale past a demo. Production teams need the same things they need from human contributors:

  • Least privilege — an agent doing code review shouldn't be able to git push.
  • Approval gates — a deploy step should pause for a human before kubectl apply runs.
  • Auditability — every run is tied to an organization, a project, a workflow instance, and a stage — you can always answer "who (or what) changed this, and under what authority."
  • Reuse — the same "Backend Engineer" agent profile, the same "commit-guard" hook, the same permission preset should work across every project in the org, not be copy-pasted into every workflow.

AgentMesh's data model exists to make those four things structural rather than best-effort.

Where to go next​

  • New to AgentMesh? Start with Core Concepts to learn the vocabulary (Organization, Project, Registry Resource, Agent Profile, Workflow, Run).
  • Ready to set something up? Jump straight to the Guides for step-by-step setup instructions.
  • Looking up an exact field or API shape? Check the Reference section.