# AI Agent Harnesses

**Point-in-time snapshot: April 2026**

A curated list of projects describing themselves as AI agent harnesses or related to "harness engineering."

## A Note On Terminology

The term **"agent harness"** emerged in late 2025 / early 2026 as a way to describe the orchestration layer that *runs* an AI agent. A harness manages the tool-dispatch loop, permissions, context, and lifecycle of an agent — as distinct from a *framework* (which provides higher-level abstractions for *building* agents) or a *backend* (which provides the inference and model serving).

The term gained traction partly through the adjacent concept of **"harness engineering"** — the practice of configuring and optimizing these orchestration layers (writing CLAUDE.md files, designing tool schemas, structuring agent prompts, etc.).

This list captures projects that self-identify with the "harness" label as of April 2026. The terminology is still fluid and not universally agreed upon — some projects listed here blur the line between harness, framework, and agent platform.

---

## Agent Harnesses

Projects that provide runtime orchestration for AI agents — tool dispatch, conversation loops, context management, and agent lifecycle.

| Project | Description | Language | Stars |
|---------|-------------|----------|-------|
| [bytedance/deer-flow](https://github.com/bytedance/deer-flow) | Open-source long-horizon SuperAgent harness that researches, codes, and creates. Handles tasks from minutes to hours using sandboxes, memories, tools, subagents. | Python | 57,759 |
| [code-yeongyu/oh-my-openagent](https://github.com/code-yeongyu/oh-my-openagent) | Agent harness (previously oh-my-opencode) with multi-provider support. | TypeScript | 48,154 |
| [langchain-ai/deepagents](https://github.com/langchain-ai/deepagents) | Agent harness built with LangChain and LangGraph. Planning tool, filesystem backend, and subagent spawning. | Python | 19,118 |
| [aden-hive/hive](https://github.com/aden-hive/hive) | Outcome-driven agent development framework and runtime harness. Self-improving, human-in-the-loop. | Python | 10,038 |
| [xerrors/Yuxi](https://github.com/xerrors/Yuxi) | Agent harness integrating LightRAG knowledge base and knowledge graphs. Built with LangChain + Vue + FastAPI, supports MCP. | Python | 4,805 |
| [mindfold-ai/Trellis](https://github.com/mindfold-ai/Trellis) | Agent harness for agentic coding workflows. | Python | 4,610 |
| [HKUDS/OpenHarness](https://github.com/HKUDS/OpenHarness) | Open Agent Harness. | Python | 3,363 |
| [can1357/oh-my-pi](https://github.com/can1357/oh-my-pi) | AI coding agent for the terminal — hash-anchored edits, optimized tool harness, LSP, subagents. | TypeScript | 2,646 |
| [kevinrgu/autoagent](https://github.com/kevinrgu/autoagent) | Autonomous harness engineering. | Python | 2,084 |
| [chrysb/alphaclaw](https://github.com/chrysb/alphaclaw) | Setup harness for OpenClaw. Deploy in minutes, stay running for months. | JavaScript | 943 |
| [e10nMa2k/cc-mini](https://github.com/e10nMa2k/cc-mini) | Ultra-light harness scaffolding for AI agents, a mini version of Claude Code. | Python | 677 |
| [truffle-ai/dexto](https://github.com/truffle-ai/dexto) | Coding agent and general agent harness for building and orchestrating agentic applications. | TypeScript | 608 |
| [Chachamaru127/claude-code-harness](https://github.com/Chachamaru127/claude-code-harness) | Claude Code dedicated development harness — autonomous Plan-Work-Review cycle. | Shell | 420 |
| [aristoteleo/PantheonOS](https://github.com/aristoteleo/PantheonOS) | Evolvable, distributed agent framework and harness for data science and bioinformatics. | Python | 370 |
| [MaxGfeller/open-harness](https://github.com/MaxGfeller/open-harness) | Code-first, composable SDK to build powerful AI agents. | TypeScript | 337 |
| [vortezwohl/Autono](https://github.com/vortezwohl/Autono) | ReAct-based highly robust autonomous agent (harness) framework. MCP support. | Python | 209 |
| [aiming-lab/AutoHarness](https://github.com/aiming-lab/AutoHarness) | Automated harness engineering for AI agents — audit, governance, safety. | Python | 181 |
| [dust-tt/srchd](https://github.com/dust-tt/srchd) | Research agents harness. | TypeScript | 69 |
| [thu-nmrc/OpenHarness](https://github.com/thu-nmrc/OpenHarness) | Fully autonomous AI agent execution framework for OpenClaw built on harness engineering. From Tsinghua University. | Python | 62 |
| [thu-nmrc/OpenHarness-For-Codex](https://github.com/thu-nmrc/OpenHarness-For-Codex) | Specialized variant of OpenHarness designed for AI-driven software development. | Python | 39 |
| [mastersof-ai/harness](https://github.com/mastersof-ai/harness) | Agent runtime with full system prompt control. Define agents in markdown, run in terminal TUI or web UI. Powered by Claude Agent SDK. | TypeScript | 31 |

## Resource Lists & Guides

Projects focused on curating resources, guides, and best practices for harness engineering rather than providing a harness runtime.

| Project | Description | Language | Stars |
|---------|-------------|----------|-------|
| [walkinglabs/awesome-harness-engineering](https://github.com/walkinglabs/awesome-harness-engineering) | Awesome tools and guides for harness engineering. | — | 1,264 |
| [AutoJunjie/awesome-agent-harness](https://github.com/AutoJunjie/awesome-agent-harness) | Curated list of agent harness projects, orchestration tools, and developer resources. | — | 277 |
| [Picrew/awesome-agent-harness](https://github.com/Picrew/awesome-agent-harness) | Awesome list of agent harness engineering resources — projects, tools, benchmarks, and guides. | Python | 109 |
| [alchemiststudiosDOTai/harness-engineering](https://github.com/alchemiststudiosDOTai/harness-engineering) | Harness engineering discussion — shortcuts, automation, hacks, and productivity with code agents. | Shell | 87 |
| [jrenaldi79/harness-engineering](https://github.com/jrenaldi79/harness-engineering) | Context engineering for coding agents — CLAUDE.md templates, enforcement, and best practices field guide. | JavaScript | 55 |

---

*Star counts as of April 2026. This is a static snapshot and will not be updated.*