hkuds/openharness

★ 12,469PythonAudience · developerComplexity · 4/5Setup · moderate

Mindmap

mindmap
  root((repo))
    What it does
      AI agent framework
      40 plus built-in tools
      Permission controls
      Context compression
    Tech Stack
      Python
      LLM backends
    Use Cases
      Build custom agents
      Personal assistant
      Agent research
    Audience
      Developers
      Researchers
    Integrations
      Slack
      Discord
      Telegram
      Feishu

mindmap root((repo)) What it does AI agent framework 40 plus built-in tools Permission controls Context compression Tech Stack Python LLM backends Use Cases Build custom agents Personal assistant Agent research Audience Developers Researchers Integrations Slack Discord Telegram Feishu

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Build a custom AI agent with 40+ built-in tools like file operations, web search, and shell commands.

USE CASE 2

Add ohmo, a personal AI assistant, to Slack or Discord to write code and open pull requests automatically.

USE CASE 3

Create an agent that asks for human approval before taking sensitive actions like deleting files.

USE CASE 4

Study a production-grade AI agent architecture for research on how harnesses, memory, and governance work.

Tech stack

Python

Getting it running

Difficulty · moderate Time to first run · 30min

Requires a Claude Code or Codex subscription to power the ohmo assistant, no separate API key needed if you already have one.

In plain English

OpenHarness is an open source Python framework for building and running AI agents, published by researchers at Hong Kong University. The core idea is that an AI language model on its own is just a text predictor. To turn it into a functional agent that can actually do things, you need a surrounding system, called a harness, that gives it tools to use, memory to store information between steps, rules about what it is and is not allowed to do, and the ability to coordinate with other agents. OpenHarness provides all of that infrastructure as reusable code. The project ships with over 40 built-in tools covering file operations, shell commands, web search, and browser interaction. It also includes a permission and governance layer so developers can set boundaries on what the agent can do, with approval dialogs for sensitive actions. Context management features let the agent remember things across long sessions and compress older context automatically when it runs long. Built on top of OpenHarness is a personal assistant called ohmo. Rather than being a basic chatbot, ohmo is designed to complete multi-step software tasks on its own. You can message ohmo through Feishu, Slack, Telegram, or Discord, and it will independently write code, run tests, and open pull requests. It works with an existing Claude Code or Codex subscription, so no separate API key is required. The project is intended both for researchers who want to study how production AI agents are structured and for builders who want a starting point for their own specialized agents. The README covers a five-minute quick-start path and documents the full harness architecture in detail.

Copy-paste prompts

Prompt 1

Using the OpenHarness framework, build me a Python AI agent that watches a folder for new files, processes each one, and sends a notification when done.

Prompt 2

Set up the ohmo assistant from OpenHarness on my Discord server and configure it to handle code reviews on a GitHub repo.

Prompt 3

Show me how to add a custom tool to an OpenHarness agent that calls my company's internal API.

Prompt 4

Using OpenHarness, create an agent with a permission layer that requires my approval before running any shell commands.

Prompt 5

Walk me through the OpenHarness context management system and explain how it compresses memory during long agent sessions.

Open on GitHub → Explain another repo

← hkuds on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.