No model lock-in: switch Anthropic, OpenAI, Gemini or DeepSeek mid-session. The agent loop, tools, MCP, hooks and permissions live in a transparent harness that evolves with your work.
npm install -g oharness
Anthropic, OpenAI and Google Gemini natively, every OpenAI-compatible endpoint, and runtimes on your own machine. Each is catalogued with its own endpoint and key, and the model list is asked of the vendor rather than shipped inside the package — so it is never a stale copy.
Define a tool with zod. JSON Schema is generated and arguments validated for you, and whether calls run in parallel or in sequence comes from the tool's own declaration.
A JSON-RPC client built in — stdio and Streamable HTTP, no MCP SDK. Servers are discovered and their tools namespaced automatically.
Compaction that cuts at a safe point rather than mid-exchange, sessions persisted and resumable, and recall across transcripts already archived.
Deny beats allow, and even bypass. Five modes, rules that fail at parse time instead of at 3am, and a prompt that is injected — so a scheduled run never waits on a human.
Delegate to isolated contexts, run them in parallel, and keep the conclusion instead of the transcript that produced it.
Connect a vendor with a key, or point at Ollama, LM Studio or your own vLLM server. Models are discovered from each vendor's own endpoint, so the list is never a stale copy shipped inside the package.
Read the docs
OHarness keeps tools, permissions, memory, artifacts and task history visible, locally owned and independent from the model provider.
Yes. The harness is free to use with your own provider key or a local model, and needs no account. The optional hosted plan bills model usage in credits.
Not yet. The npm packages are published under the MIT licence; the source repository is private. Open source: coming soon, with no release date announced.
Yes, natively — along with DeepSeek, Kimi, GLM, Qwen, OpenRouter and 30+ other providers, and any OpenAI-compatible endpoint.
Yes. Ollama, LM Studio and vLLM are built in and need no key, so model requests never leave your machine.
Yes. /model switches mid-session. Provider-neutral history, tool results, permissions, memory and artifacts stay with the harness.
Not with your own key: requests go straight from your machine to the provider you chose, and keys are stored locally with owner-only permissions. Only the optional hosted plan routes requests through our metering gateway.
Claude Code is built for Anthropic's models. OHarness follows the same architecture — agent loop, tools, permissions, MCP, subagents — with the model as a replaceable backend, plus browser and desktop frontends on the same engine.
The CLI runs anywhere Node 22.6 or newer does: macOS, Linux and Windows. The desktop app is published for macOS on Apple silicon.