Choose your path

Pick Service, Harness, or Harness UI by where your agents run and who operates them.

Agent Foundation supports managed execution through Service, embedded execution through Harness, and interactive work through Harness UI.

Your goalWhat you needStart here
Deploy a shared agent platformDocker Compose and a model provider account for the local trialService quickstart
Use your team's agentsConsole URL, an account, and permission to run an agentUse an existing platform
Call remote agents from an applicationService URL, workspace API key, and a configured agentConnect your application
Run agents inside your Python processPython 3.13+ and uv; the first example uses an offline modelHarness quickstart
Work interactively in a repositoryHarness UI and a supported model subscription or API keyHarness UI setup

Choose where execution lives

  • Service: your application submits work to a running deployment. Service owns identities, saved conversations, and durable execution; Console is its browser application.
  • Harness: your Python process constructs and runs the agent. Your application owns storage, access policy, and delivery.
  • Harness UI: the terminal and browser workbench runs agents with its own configuration and conversation history. Share an instance with trusted collaborators.

A Service SDK is a client for remote execution. The Harness library executes agents inside your process. Service Console and Harness UI's browser are separate applications.

Use individual components

NeedGuide
Portable files and commands, with or without an agentEnvironments
Environment operations through a daemonEnvd
AG-UI events from Harness observationsStream Protocol
Distribution names, imports, and runnable examplesPackage catalog

For a guided explanation of one conversation, read Core concepts. For deployment operations, see Run and maintain. Service clients have independent versions; check their supported contract against your deployment.

On this page