Finally, an OpenRouter for agent harnesses!

HarnessRouter is an infrastructure layer that transforms existing harnesses such as Codex and Claude Code into plug-and-play agent backends. This enables products to run tasks, retrieve results, and switch harnesses using a single API. The system adheres to the Unified Harness Protocol (UHP) and offers an OpenAI Responses-compatible API.

It manages persistent sessions, streaming progress, files, artifacts, cancellation, and structured failures. Eight harnesses integrated with various models can operate simultaneously on a single task. Each harness configuration provides a comparison of the best and worst results.

To begin, users can run HarnessRouter locally or integrate it into their products. Starter kits, managed Cloud Self-host Community Edition, and Docker commands facilitate the setup process. Docker requires approximately 4 GB of disk space and a provider API key, while no HarnessRouter account or bundled model or trial key is necessary. The named volume maintains the database, files, installed harness CLIs, and workspaces between restarts.

Users can create custom harnesses for specific needs or use built-in harnesses without any additional steps. Once a harness is selected, users can initiate a task, monitor progress in real-time, and access the files generated by the agent. Custom harnesses allow for tailored behavior specific to a product. The agent harnesses serve as pluggable agent backends, enabling users to call the self-hosted instance's OpenAI Responses-compatible API and select the desired harness using the metadata.harness_id.

The API documentation can be accessed at $HARNESSROUTER_BASE_URL/v1/openapi.json, with a browsable copy available at /v1/docs. The default URL is applicable when the backend and customer environment (CE) run on the same computer. For remote instances, a reachable URL for the CE should be used. The self-hosted API and networking guide provide further details.

HarnessRouter empowers developers to incorporate advanced agentic features into their products, such as content creation and data analysis, using harnesses as seamless and pluggable backends.