Getting started

Every Agent Mesh has a control plane (which enrolls members, distributes policy, and runs the router) and the members (agentmesh-node or native SDK programs) that connect to it.

To connect an agent or machine to a mesh, you need a running control plane and its URL (https://...).

How to get a control plane and its URL

Pick the option that matches what you want to do:

Control plane optionHow to run it and obtain its URLArchitecture & IngressBest for
1. Shared Public TestnetAlready running — use https://bananas.sam-mesh.devHosted shared control plane and routerTrying agentmesh-node and calling your first remote tool or model in 60 seconds (Quick start).
2. Local Control Plane (agentmesh-one)Run agentmesh-one --data-dir ~/agentmesh-one --tunnel cloudflare --tunnel-install and copy API URL: from the startup bannerSingle binary (control plane + router + console) with SQLite and an HTTPS tunnelRunning your own private mesh from a workstation or VM in seconds (Your own mesh).
3. Codespace Control Plane (agentmesh-one)Open the repository in a GitHub codespace, run make testnet, make port 8080 publicagentmesh-one in a container on your GitHub account, published on https://<codespace>-8080.app.github.devA control plane of your own with nothing installed and no cloud account, for as long as the codespace runs (Codespaces).
4. Cloud Control Plane (agentmesh-one)Cloud Run: gcloud run deploy agentmesh-one --image ghcr.io/google/agentmesh-one:latest ...
SkyPilot: sky launch -c agentmesh-hub deploy/skypilot/agentmesh-one.yaml
Always-on agentmesh-one with managed TLS/WSS ingress + PostgreSQL or persistent disk (can also run on cloud free tiers for testing)Operating a dedicated production control plane in your own cloud account (Cloud Run · SkyPilot).

Walkthroughs

  1. Quick start: Join the public bananas.sam-mesh.dev testnet, discover remote MCP tools and inference models, and give your AI agent the Agent Mesh skill.
  2. Your own mesh: Start agentmesh-one in one command, grab the Control Plane URL from the startup banner, publish a local Ollama model, and call it from another node.