HarnessRouter

What is HarnessRouter?

HarnessRouter is a unified API for running agent harnesses like Codex, Claude Code, Hermes, and DeepSeek Harness inside your own product. Each task gets an isolated cloud sandbox, while the platform handles sessions, streaming, files, resume, and traces. Teams switch harnesses or models by configuration, or self-host the open-source edition.

Top Features:

• One API: reach Codex, Claude Code, Hermes, and more through one request format.
• Isolated sandboxes: every task runs in its own hosted sandbox that scales automatically.
• Harness Arena: compare harness and model pairs on cost, speed, and accuracy.

Use Cases:

• Agent features: add a coding or research agent to a SaaS product.
• Backend migration: move an existing agent backend over without changing your product API.
• Cost cutting: route each workload to the cheapest harness and model that works.

Who Can Use HarnessRouter?

• Product teams: ship slides, sheets, dashboard, or video agents inside existing apps.
• Backend engineers: skip building sessions, streaming, files, and traces around each harness.
• AI startups: test several agent runtimes before sending production traffic to one.

Pricing

• Free ($0): 30 agent work minutes, one workspace, and seven day retention.
• Developer ($20/month): $20 of monthly usage included, plus 30 percent off top-ups.
• Production ($200/month): $200 of usage, five workspaces, and 50 percent off top-ups.
• Scale ($1,000/month): $1,000 of usage, unlimited workspaces, tracing, and bigger top-up discounts.

Pros and Cons

Pros:

• No lock-in: switching harnesses or models is a configuration change, not a rebuild.
• Open source core: the Apache 2.0 edition keeps data on your own infrastructure.
• Pay for work: usage is drawn only while tasks run, so waiting is free.

Cons:

• Developer focused: setup needs API work, so non-technical teams need engineering help.
• Uneven features: capabilities differ by harness, so results can change after switching.
• Young product: the platform and its protocol are new, with little public history.

FAQs:

1) What is an agent harness?

It wraps a model with a sandbox, tools, and a working loop.

2) Which harnesses are supported?

Codex, Claude Code, Hermes, Pi, Qwen Code, and DeepSeek Harness, among others.

3) Can I bring my own keys?

Yes, your provider bills model usage directly and you pay for agent work.

4) Can I self-host it?

Yes, the Apache 2.0 Community Edition runs on infrastructure you control.

5) Is there a free plan?

Yes, it needs no card and includes 30 minutes of agent work.