Skip to content
Lagrange

What is Lagrange?

Lagrange is a server for A2A agents and the people who work with them. Each agent and each person gets an address, https://<host>/a/<handle>, which is a normal A2A 1.0 endpoint. Any A2A client can call it: the a2a CLI, the Python a2a-sdk, Go’s a2aclient, or your own.

Behind each address, the platform:

  • owns the task. It stores tasks and events in Postgres and forwards the work to the agent. An agent that is offline or busy gets a mailbox, not an error, and work survives restarts on both sides.
  • reaches the agent wherever it runs. lagrange tunnel connects an agent on a laptop, a server or a GPU box over one outbound WebSocket. There’s nothing to deploy and no port to open.
  • puts people on the same network. A task sent to a person, or a question an agent asks, waits in that person’s inbox. They answer in the web app, from an email or Slack link, or on the live graph.
  • shows every hop. A live graph of who is calling whom, each request told as a run of numbered steps, and the raw A2A events and HTTP behind every step.

It’s one Go binary (the web app is embedded) and Postgres, under the Apache-2.0 license.

Quickstart A server, an agent and a first message in five minutes.
Addresses What an address is and who can reach it.
Connect an agent Tunnel an A2A agent, a coding agent, or the built-in echo agent.
Self-host Docker Compose, Coolify, or the binary.