Breaking

HarnessRouter Open-Sources a Unified API for Running Codex, Claude Code, Hermes and Other Agent Harnesses

September 22, 2026 at 06:46 EDT

AI Agents
Software Dev & Coding
Open Source

HarnessRouter has released an open-source Community Edition that lets developers run multiple AI agent harnesses through one interface, positioning the project as a routing layer for agent execution rather than model inference. The Apache 2.0 package supports 14 harnesses, including Codex, Claude Code, Hermes, Pi, DeepSeek Harness, OpenCode, Qwen Code, Cline, Gemini CLI, goose, Aider and OpenHands, behind a shared Agent API.

Sep. 2026 · HarnessRouter

One API for agent harnesses

HarnessRouter Community Edition brings Codex, Claude Code, Hermes, DeepSeek Harness and other agent runtimes behind a shared execution interface for sessions, streaming, files and artifacts.

supported harnesses in Community Edition

benchmark configurations on one task

≈475x

reported cost range in the test

3.2x

reported latency range

From one integration to 14 harness choices

API surface

harness backends

The core proposition is routing complete agent runtimes, not only individual model calls.

Why teams care

A shared interface can reduce integration work and make harness cost, latency and quality comparisons more repeatable.

What remains open

Benchmark results are task-specific, and self-hosting still requires Docker, provider keys and operational ownership.

Execution layer, not a model router

Task→Harness choice→Tools and sessions→Artifacts

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.