HarnessRouter Open source

What it does

HarnessRouter Community Edition is a self-hosted infrastructure layer that puts a single, unified API in front of multiple AI coding-agent harnesses — Codex, Claude Code, Hermes, PI, DSH, and others. Instead of writing separate integration code for each agent CLI, you talk to one Gateway that implements the Unified Harness Protocol (UHP) and exposes an OpenAI Responses API-compatible interface, so existing SDKs and tooling work with minimal changes.

What it gives your stack

The project ships as three Docker services: a Console (web UI for configuration and task management), a Gateway (the REST API), and a Runner that executes each agent harness inside an isolated, per-session OS user and filesystem workspace rather than a shared container. It handles persistent sessions with streaming progress, file upload/download inside agent workspaces, task cancellation, structured error reporting, environment-variable and secret management with automatic redaction, and bring-your-own-key multi-provider support. A handful of starter kits (slide generation, spreadsheet automation, SQL dashboards, video creation, a workplace collaboration demo) show the pattern of building a product on top of a pluggable agent backend.

Works with

Codex, Claude Code, and other harnesses the Runner supports (Hermes, PI, DeepSeek Harness, DSH) through the Unified Harness Protocol; any OpenAI Responses API client can talk to the Gateway.

How to install or connect

docker run -d --name harnessrouter -p 127.0.0.1:3000:3000 -v harnessrouter:/data harnessrouter/harnessrouter

Default login is harnessrouter/harnessrouter — change it immediately after first boot. The stack needs roughly 4GB of disk for the persistent data volume.

Maintenance and safety

Actively developed (600+ commits on main, ~2,800 GitHub stars, 289 forks, a handful of open issues and PRs) and licensed Apache-2.0 for the Community Edition; upstream harness CLIs keep their own licenses and the starter kits carry separate terms. Running untrusted agent sessions still means reviewing what each harness can touch on disk, since the Runner's isolation is OS-user-level rather than full containers.

Who should use it

Teams building a product or internal tool on top of more than one coding-agent CLI — or who want to self-host that layer instead of depending on a single vendor's hosted API — rather than individual developers just using one agent directly.