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

The world's first unified interface for agent harnessesBacked by Y Combinator

Bring the world's best AI agents into your app, with one API.

Run Codex, Claude Code, Hermes, and more in isolated sandboxes as your product backend. Ship faster, scale reliably, and save costs.

Start building free

Plus 500 free credits at launch. Limited time.

Launch Partner:

Used by

The old way

The problem

Building AI agent backend for takes months.

Your backlog, if you build it yourself.
  • sandbox per run
  • agent runtime
  • tool orchestration
  • files & artifacts
  • sessions & streaming
  • retries & timeouts
  • permissions
  • cost controls

And the upgrades, fixes, and maintenance never stop.

The new way

The solution

We run the AI agent backend. You ship it in minutes.

One API brings leading AI agents into your app.
  • Codex
  • Claude Code
  • Hermes

We take care of the upgrades, fixes, and maintenance for you.

Skip the backlog

The steps

Build it in three steps.

codex · claude code
1 · drop it into your coding agent
“Build a founder launch-video app: users drop a logo and an idea, get a launch video.”
2 · describe your idea
server envHR_KEY=sk-hr-••••••••
3 · add your API key securely
SHIP

Your user gets work, not words. Start with one harness-powered feature.

Start building freeLearn by watching demo video

Build demo

See it in action.

A Cursor-style coding app, built with the same three steps. Watch the full run from AGENTS.md to a working product.

Build your next unicorn product
Build demo · Cursor-style coding app0:54

Use cases

Some AI features your app can ship, and more.

Every one of these is a real request-to-artifact workflow: your user asks inside your product, a harness does the work, and the finished result comes back.

Your appSite builder
HarnessRouterOne API
  • CodexCode, apps and images
  • Claude CodeCode, files and documents
  • HermesAutonomous, any frontier model
  • PiSoonCoding agent
    Soon

Every harness ready, one API

Build your first use case

Scale

Scale without operating the infrastructure.

Run concurrent agent workloads in isolated sandboxes. HarnessRouter provisions and operates the serverless execution infrastructure for you.

Your appOne APIServerless agent runtime
5 active sessions
Codex2 sessions
Session C-01Running
Isolated sandbox
Session C-02Running
Isolated sandbox
Claude Code2 sessions
Session A-01Done
Isolated sandbox
Session A-02Running
Isolated sandbox
Hermes2 sessions
Session H-01Starting
Isolated sandbox
Session H-02Running
Isolated sandbox

Every session runs in its own isolated sandbox. HarnessRouter provisions the runtime on demand, streams events as the run progresses, and returns the finished artifact to your app.

Parallel sessionsIsolated by sessionCreated on demand
Scale with serverless sandboxes

Optimize

Improve performance and control costs.

Trace and compare harness × model combinations across quality, cost, and latency to identify the best setup for each workload.

HarnessRouter

Same task, same input · Harness × model comparisons

Task benchmark

Completed4 configurationsSame task, same inputCost captured
Harness / modelSuccess rateLatencyCost
HermesProduction pickgpt-5.2100.0%2m 33s0.47 credits
Codexgpt-5.2100.0%4m 36s0.72 credits
Hermesclaude-opus-4.860.0%2m 42s150 credits
Claude Codeclaude-opus-4.860.0%3m 18s223 credits
Selected: Hermes × gpt-5.2Winner for this workload: lowest-cost 100% successful run
HarnessRouter benchmark · Same task, same input475× difference in cost

The lowest-cost successful run saved 99.8% in credits against the costliest configuration, on the same task with the same input. Scores and success rates differ by setup, and results vary by workload.

Optimize production costsView the benchmark
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

Q&A

The question everyone asks first.

What exactly is a harness?

An LLM returns tokens. A harness gives it a sandbox, tools, and a loop, so it returns the actual file. GPT-5.2 and Claude Opus 4.8 are LLMs. Codex and Claude Code are harnesses. A harness also carries the agent's configuration: its instructions, which model it runs, and what it is allowed to do.

Why not just use Codex or Claude directly?

You should. Use them to build your app or tools. But they work for you, at your desk, not for your users. HarnessRouter puts them to work for your users, inside your product. Your app sends a task through one API and gets back finished work: videos, games, codebases, docs, and more. Your users never touch a terminal.

How much does it cost?

Create an account, add a card, and get 500 free credits. When you need more capacity, choose Developer, Production, or Scale. Additional usage can draw from purchased top-up credits.

Do I need to understand agents to use this?

No. Drop our AGENTS.md into Cursor or Claude Code, describe your feature, and your coding agent wires everything: the endpoint, the config, the UI. The demos above were built exactly this way.

Am I locked into one harness?

The opposite. That is the point. Your config names the harness: claude-code today, codex tomorrow, or back again, swapped anytime with one line. Your integration code and output schema never change. When a better harness ships next quarter, your product gets better with a config change.

What exactly comes back to my app?

Structured, renderable results: code changes as reviewable diffs, generated files and documents, images, and confirmations of real tool actions like GitHub, Slack, Notion, or internal APIs. Your UI decides how to show them and when the user accepts.

Put agent harnesses to work in your product.

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

Start building free