Agent Foundation

The open-source, self-hosted foundation for AI agents. Deploy the managed Service, embed the Harness library in your application, or work interactively with Harness UI.

Choose your path

Products

Service

Teams and applications

A self-hosted platform with identities, workspaces, durable execution, and saved conversations, used through Console or the HTTP API.

Get started Overview

Harness

Python developers

A library that composes agents from instructions, tools, and capabilities and runs them in your process. Your application owns storage and access.

Get started Overview

Harness UI

Individuals and small teams

A terminal and browser workbench for working with agents in your repositories, with shared conversations, files, and terminals.

Get started Overview

Start with a task

Deploy a shared agent platformRun Service locally and configure a model Use your team's agentsStart a conversation on an existing platform Add managed agents to your applicationAuthenticate, submit work, and read results Build an agent in PythonRun an offline example, then add tools and models Work interactively in a repositoryConnect a model and use the terminal or browser Give an application file and command accessUse the same providers with or without an agent

Building blocks

EnvironmentsPortable file, command, process, and computer access EnvdFile and process operations over the Environment Interaction Protocol Stream ProtocolTyped AG-UI events from Harness observations LoggingStructured logs and process-wide logging setup

This site tracks main. Agent Foundation is in active 0.x development; APIs and configuration may change between minor releases.

See Contributing for source setup, the specifications for architecture contracts, and the package catalog for distributions and examples.