OpenHarness
An open-source Python AI agent framework supports local models through Ollama, persistent memory, delegated agents and permission controls.
OpenHarness is a Python framework for developers and researchers building AI agents that can act on files, use tools and retain work across sessions. It includes ohmo, a personal agent that handles coding tasks through Feishu, Slack, Telegram or Discord.
- Coding and tool use: Agents can read and edit files, execute shell commands, search code and the web, and call MCP tools; ohmo can also create branches, run tests and open pull requests.
- Persistent work: Memory, session history and resumable sessions keep context available between conversations, while automatic compression supports long sessions.
- Agent coordination: Delegate work to subagents and manage shared tasks or background jobs.
- Extensions and permissions: Add Markdown skills and compatible Anthropic plugins, and control tool access with path rules, command rules and interactive approvals.
OpenHarness is open source under the MIT license. It supports local LLMs through Ollama, plus OpenAI-compatible and Anthropic-compatible APIs for services including DeepSeek, Claude and Gemini. Local Ollama inference runs on your hardware; cloud backends send model requests to the selected service.
Ohmo can use an existing Claude Code or Codex subscription without a separate API key. OpenHarness also supports GitHub Copilot through OAuth authentication.
