The world’s first unified interface for agent harnessesOpen SourcePowered by UHP
HarnessRouter

Navigate HarnessRouter

Search the site

  • HarnessRouter Community EditionOpen-source agent harness infrastructure and the Unified Harness Protocol.Product
  • Harness catalogSupported harness IDs and source-backed integration references.Harnesses
  • Care Prep BenchmarkThe same task compared across Codex, Claude Code, and Hermes configurations.Benchmarks
  • Harness rankingsPublic attributed usage across reviewed agent harnesses.Product
  • ExamplesSee how agent-backed features return finished work inside a product.Product
  • PricingPlans and usage pricing for HarnessRouter.Product
  • Hermes AgentAn open-source, persistent AI agent with memory, reusable skills, built-in tools, browser automation, scheduled tasks, and subagents.Harnesses
↑↓ Navigate↵ OpenEsc Close
Sign up
Backed by Y Combinator

Build agent products without handling harness engineering.

Plug Codex, Claude Code, Hermes, DeepSeek Harness, and more into your product as agent backends through one API.

Start buildingBook a demo

The best agent harnesses as your infrastructure, one conversation away.

Finding the best time slots for you…

HarnessRouter Community Edition - Open-source unified interface for agent harnesses | Product Hunt

Used by teams building agent-powered products

How HarnessRouter works

One API between your product and every agent harness.

Start newBuild a new agent product

Start with a ready-to-use coding agent.

MigrateMigrate an existing agent product

Move your current managed agent backend to HarnessRouter.

Open Source at the core

Use the managed Cloud or run it yourself.

HarnessRouter Cloud

Start at $0, no card needed.

Explore the console and configure agents for free. Usage is drawn only while hosted tasks run; waiting is free.

Use a HarnessRouter key

Every model at its listed catalog price, no markup. Plans from $20 a month include their full price as usage.

Bring your own API keys

Your provider bills your model usage directly. You pay for Agent work only; waiting is free.

Start with HarnessRouterView Cloud pricing

Community Edition

Self-hosted

Run the same UHP interface on infrastructure you control.

  • Apache 2.0
  • Your infrastructure
  • Your keys, state, and files
View on GitHub

Benchmark · One task, same input

Cost, speed, and accuracy depend on the harness, not just the model.

Benchmark the complete harness × model configuration on the work your product actually runs.

Run your task in Arena
Observed cost spread1×

Lowest to highest configuration on this measured task.

Harness × modelRelative cost
View methodology

This is one measured workload, not a universal maximum. Harness and model both change production cost.

Serverless agent runtime

Scale without operating the infrastructure.

HarnessRouter provisions and operates one isolated sandbox for every task, so workloads scale without infrastructure work.

Book a demo
Concurrent tasks
  1. Tasks received
  2. Provisioning
  3. Running
Your product20 task requests
1 API
HarnessRouterProvisions and operates
Isolated sandboxesOne task per sandbox
20 active

Your first run

Build it in three steps.

Then manage your harnesses and tasks in the console.

1

Step 1 of 3

Copy and paste AGENTS.md

AGENTS.md
Copy AGENTS.md
Paste into your coding agent

Harness × model routing

Switch harnesses and models without changing your product API.

Benchmark harness × model combinations on your task, then choose the configuration you route to, all through one integration.

One contract between your product and a complete agent harness.

Your product

Task in · result out
Product chat Agent online
You

Create a cinematic car video

Agent
Routing taskHarnessRouter is running the selected agent
HarnessRouterUHP server
  1. Starts the harness + model
  2. Relays progress and events
  3. Returns results + artifacts

Choose the combination

Harness × model
Harness
Model

Benchmark the pair by quality, latency, and cost.

Read the Unified Harness Protocol →

Open Source starter kits

Start from a complete agent product.

Each Open Source kit includes an app, a configured agent, and the skill that teaches it what to build.

View all starter kits Read the docs
Slides starter kit interface
01

Slides

Build editable presentations from a prompt.

Sheets starter kit interface
02

Sheets

Run an agent across rows and fill results live.

Dashboards starter kit interface
03

Dashboards

Turn database questions into live query-backed panels.

Videos starter kit interface
04

Videos

Plan, render, arrange, and assemble generated clips.

Customer story · Readily

“We can go from a problem and a plan to an agent in product within 24 hours.”

Co-founder and CEO, Readily (YC S23)

Readily uses HarnessRouter to bring domain expertise into production agents without rebuilding the agent runtime.

Go from plan to product

Technical Q&A

Questions engineers ask before integrating.

Direct answers about the runtime, protocol, hosting, and cost.

Product fit

What is an agent harness?

An LLM returns tokens. A harness gives it a sandbox, tools, and a loop, so it returns actual work. GPT-5.2 and Claude Opus 4.8 are models. Codex and Claude Code are harnesses.

Is HarnessRouter just another AI wrapper?

No. HarnessRouter is the execution and runtime layer between your product and supported agent harnesses. Your application sends a task and receives progress, files, artifacts, final state, and trace through one interface.

Why not integrate Codex or Claude Code directly?

You can if you only need one harness and want to own the surrounding runtime. HarnessRouter becomes useful when you need multiple harnesses, or do not want to rebuild sessions, streaming, files, resume, and traces around each integration.

What backend engineering does HarnessRouter remove?

HarnessRouter handles the agent execution lifecycle: sessions, tools, files, artifacts, streaming, cancellation, errors, resume, and traces, with hosted sandboxes on Cloud.

Interface and runtime

What is UHP?

The Unified Harness Protocol is an open standard for the interface between your product and a complete agent harness. It defines how a harness is discovered, selected, and configured, including its model, skills, tools, and MCP servers, and how tasks, sessions, streaming events, files, artifacts, cancellation, errors, and results flow between the two. Any harness that implements it is interchangeable behind your product. Read the Unified Harness Protocol

Can we switch harnesses without changing our product API?

Yes. Your product keeps the same HarnessRouter integration while selecting a supported harness by configuration. Harness-specific capabilities can differ, so you can benchmark combinations before changing production traffic.

What does HarnessRouter return to our application?

Your application can receive streamed progress and events, task status, generated files and artifacts, the final result, errors, and execution trace.

Deployment and control

What is the difference between Community Edition and HarnessRouter Cloud?

Community Edition is Apache 2.0 and runs on infrastructure you control with your own keys, state, and files. HarnessRouter Cloud runs the hosted runtime for you, with sandboxes that scale on demand.

Where do our keys, files, sessions, and state live?

With Community Edition, the runtime and application state stay in your environment. Requests to the model provider still follow the provider and credentials you choose. With Cloud, HarnessRouter operates the hosted runtime for you.

Cost and adoption

How does pricing work with our own API keys or a HarnessRouter key?

Start at $0 with no card: explore the console and configure agents. Bring your own API keys and your provider bills your model usage directly; you pay for Agent work only, and waiting is free. Use a HarnessRouter key and every model is billed at its listed catalog price with no markup; plans from $20 a month include their full price as usage. See plans and pricing

What is required to migrate an existing agent workflow?

Adopt the HarnessRouter API or UHP contract, then map your existing workflow, tools, files, and result handling to it. Community Edition and starter kits shorten the path, but your team still verifies product-specific behavior.

Put agent harnesses to work in your product.

Start with one API. Scale and optimize as your workloads grow.

Start building