What is OpenHarness?
OpenHarness is an open-source Python implementation of an agent harness, the runtime infrastructure surrounding a language model. It manages tool execution, context compression, persistent memory, session history, and subagent coordination. Researchers and developers can extend it with custom tools, skills, plugins, and model providers.
It installs locally on Linux, macOS, and Windows and runs through a command-line or terminal interface. Provider profiles support compatible model APIs, local models through Ollama, and subscription bridges for Claude Code and Codex. The bundled ohmo personal agent connects to Feishu, Slack, Telegram, and Discord to perform tasks such as editing code, running tests, and opening pull requests.
Key capabilities
- Agent execution Runs streaming tool-call loops with parallel tool execution, retries, and token and cost tracking.
- Tools and extensions Provides file, shell, search, web, and MCP tools, with on-demand skills and plugins for commands, hooks, and agents.
- Context and memory Supports context compression, persistent memory, project instruction discovery, and resumable sessions.
- Permission controls Applies permission modes, path and command rules, lifecycle hooks, and interactive approval dialogs.
- Multi-agent coordination Spawns subagents, delegates work, tracks team tasks, and manages background execution.
- Provider profiles Separates model settings and credentials across compatible API endpoints and subscription workflows.
Where OpenHarness fits
OpenHarness fits developer workflows that need an agent to inspect repositories, modify files, execute tests, and coordinate delegated tasks. It also provides a foundation for specialized agents with reusable skills and custom tools. Through ohmo, these workflows can be operated from supported chat channels rather than only from a terminal.
Before deployment, teams should verify provider authentication, tool permissions, sandbox settings, and remote-channel access policies. Project skills are discovered automatically and should be disabled for untrusted repositories. The dry-run preview inspects configuration without executing tools or contacting MCP servers, so it does not replace a live connectivity and execution test.
Install
pip install openharness-aiReviews of OpenHarness
Be the first to share your experience with OpenHarness.
Write a review
Loading review form…
Alternatives to OpenHarness
LangChainLangChain is an MIT-licensed open-source framework (Python/JS) for building LLM applications and agents via LangGraph; LangSmith adds optional hosted tracing, evaluation and deployment.
CrewAICrewAI is an MIT-licensed open-source Python framework for orchestrating multi-agent AI workflows; CrewAI Enterprise adds a hosted runtime with a free Basic tier and custom-priced governance features.