Open Source Universal Agent API

One API for Every AI Agent

Write your agent code once and run it on any harness — Anthropic SDK, Goose, LangChain, Letta, and more. Stop rewriting integrations. Start building.

View on GitHub

The Problem

Every AI agent harness has its own API, its own patterns, its own quirks.

  • Rewrite integration code for each harness
  • Learn different APIs, types, and conventions
  • Vendor lock-in makes switching painful
  • No way to compare harnesses objectively

The Solution

Open Harness provides a single, unified API that works across all major agent harnesses.

  • Write once, run on any supported harness
  • Consistent API for execution, tools, memory
  • Switch harnesses with a one-line change
  • Compatibility matrix shows real capabilities

Get Started in Seconds

Python TypeScript
from openharness_anthropic_agent import AnthropicAgentAdapter from openharness.types import ExecuteRequest # Create adapter - swap this line to switch harnesses adapter = AnthropicAgentAdapter(api_key="...") # Same code works with any adapter async for event in adapter.execute_stream(ExecuteRequest(message="Hello!")): if event.type == "text": print(event.content, end="")

Available Adapters

Production-ready adapters for the most popular AI agent harnesses

🧠 Available

Letta

Memory-first architecture with persistent memory blocks, agent lifecycle management, and self-editing capabilities.

Persistent Memory Agent CRUD Inner Monologue
🦆 Available

Goose

MCP-first architecture with native Model Context Protocol support and 25+ model providers.

MCP Native 25+ Providers Sessions
Available

Langchain Deep Agents

Sophisticated agents built on LangGraph with planning, subagent delegation, and file system operations.

Planning Subagents File Ops
💻 Available

Claude Code

Full-featured CLI agent with skills, MCP integration, hooks, subagents, and advanced file operations.

Skills MCP Hooks Subagents

Documentation

Everything you need to integrate and extend Open Harness

📖

API Reference

Complete specification covering execution, tools, memory, agents, sessions, and more.

Conformance Results

See test results and capability coverage for each adapter.

📄

MAPI Specification

Machine-readable API specification in Markdown API format.

Ready to unify your agent integrations?

Join the growing community building portable AI agent applications.

Star on GitHub Get Started