OpenHarness
A Python toolkit for building AI agents, plus ohmo, a personal agent you can chat with in messaging apps.
- Stars
- 16.2k
- Forks
- 2.6k
- Licence
- MIT
- Language
- Python
- Last update
- 5 months ago
- Type
- Dev tool
What it is
OpenHarness is an open-source Python project that provides the supporting infrastructure around a language model so it can act as an agent: calling tools, using skills, keeping memory, and coordinating sub-agents. Its command-line entry point is called oh. ohmo is a personal agent built on OpenHarness that the README says can be reached through Feishu, Slack, Telegram, or Discord, and that it can run on an existing Claude Code or Codex subscription.
- Run the oh command to start an agent session in your terminal
- Use built-in tools for file editing, shell commands, search, and web access
- Load custom skills written as Markdown files on demand
- Chat with ohmo through a supported messaging app, according to the README
- Preview runtime settings and readiness with oh --dry-run before running a prompt
Great for
- Developers who want to see how an agent harness is structured in readable Python
- Researchers experimenting with agent tools, skills, and coordination patterns
- Builders who want to extend an agent with custom plugins or providers
- People who want a personal agent reachable from chat apps, if they already have a supported subscription
Not ideal for
- Non-technical users looking for a polished consumer app, since setup and configuration are developer-oriented
- People who need a stable production system, since the latest release is v0.1.9 and the README's What's New section lists many recent changes
- Anyone uncomfortable giving an agent shell, file, and web access, since the README describes these as core capabilities that need permission controls to manage
How to get it
- 1
Install with a package manager
You'll need: Python 3.10 or newer
Clone the repository
git clone https://github.com/HKUDS/OpenHarness - 2
Run it
- Install Python 3.10 or newer
- Clone the repository and follow the Quick Start section of the README
- Run oh to launch the agent in your terminal
- Try oh --dry-run to check your configuration before running a prompt
- 3
Get the most out of it
- Use oh --dry-run to check auth, MCP config, and skills before a real run
- Add project context through CLAUDE.md files, which the README says are discovered and injected
- Use MEMORY.md for persistent memory across sessions, as described in the README
- Write custom skills as Markdown files and load them on demand
- Use permission modes and PreToolUse/PostToolUse hooks to control tool behavior
Downloads
Strengths
- Compact, readable Python codebase focused on the agent loop, tools, and memory, according to the README
- Includes permission modes, path and command rules, and hooks for controlling what the agent can do
- MIT license with an active repository; the last push is dated 2026-06-04
Trade-offs
- The README is partly promotional and some claims, such as a 'personal agent that works for you', are not independently verified here
- ohmo depends on an existing Claude Code or Codex subscription, which the README says removes the need for a separate API key but still requires that account
- Many features are new and the README lists a Roadmap item (ClawTeam integration), so some capabilities may still be changing
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| OpenHarness (this) | 16.2k | Dev tool | intermediate | 5 months ago | Developers who want to see how an agent harness is structured in readable Python |
| openclaw | 394k | App | intermediate | 4 days ago | People who want an AI assistant that runs on their own machine rather than a hosted service |
| nanobot | 49.8k | App | intermediate | 4 days ago | People who want a self-hosted personal assistant they can inspect and customize |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.