One API for multiple harnesses
Integrate once and select the harness for each task. HarnessRouter presents a shared contract for Codex, Claude Code, and Hermes so teams can switch harnesses without rebuilding their backend integration.
HarnessRouter 是一个 Agent 后端层,让团队能在自己的基础设施上通过单一 API 将 Codex、Claude Code 和 Hermes 接入产品。对于不想自行运行 Agent Runtime 的团队,也提供托管云端选项。
HarnessRouter is an agent infrastructure product for teams building AI features into software products. It provides one Agent API for running supported harnesses such as Codex, Claude Code, and Hermes, while keeping the product’s own UI, authentication, and data model in the customer’s application.
The open-source Community Edition runs on infrastructure you control and packages the Gateway, Runner, and Console in one Docker container. The same product line also includes HarnessRouter Cloud for teams that want the API without operating the backend themselves. The homepage and docs emphasize production agent workflows such as sessions, streaming, files, artifacts, cancellation, and structured failure handling.
The product is positioned as a backend layer for completed work rather than a simple model router. The docs and guides focus on product integration patterns, server-side authorization, multi-tenant behavior, and the workflow needed when users expect agents to return finished outputs instead of only model responses.
Integrate once and select the harness for each task. HarnessRouter presents a shared contract for Codex, Claude Code, and Hermes so teams can switch harnesses without rebuilding their backend integration.
The open-source edition runs in a single Docker container on infrastructure you control. The page says no managed database, external vault, or cloud account is required, and that the API, console, SQLite state, and workspace files stay on your own infrastructure.
The docs list sessions, response streaming, continuing a session, and canceling a session as core API capabilities. This supports long-running agent work where users need progress updates and the ability to resume or stop tasks.
HarnessRouter supports uploading files, persisting workspace files, and returning finished artifacts. That makes it suitable for product flows that need input documents, generated outputs, and downloadable results.
UHP is the public, versioned contract behind the platform. The homepage describes it as an open protocol with OpenAPI 3.1, JSON Schema 2020-12, versioned API definitions, and runnable checks that can be verified locally.
The Starter Kit provides complete examples that show how sessions, streaming, files, and finished work fit into real product experiences. The site also highlights a Cursor-style coding app, a family-care demo, and a Slides Kit production example.
Use HarnessRouter when your product needs to launch harnessed agents behind the scenes while keeping the product UI and user experience in your own app. The docs and examples frame it as a server-side agent backend layer.
Use the session model for work that unfolds over time, such as coding tasks, document work, or other jobs that need streaming progress, follow-up actions, and later inspection.
Use the files and artifacts flow when users need to upload source material and receive generated outputs back from the agent. The homepage explicitly calls out workspace files and finished outputs as part of the lifecycle.
Use the Starter Kit when you want a concrete starting point instead of an empty repository. The site shows reference demos and production kits for teams that want to adapt an existing agent product pattern.
Use Community Edition if you want to control infrastructure yourself, or move to HarnessRouter Cloud if you want the same API without running the agent backend. The pricing page and homepage both present this split clearly.
No. The site presents Community Edition as the open-source, self-hosted option and also offers HarnessRouter Cloud for teams that want a fully managed service.
The homepage says Community Edition includes the Gateway, Runner, and Console in one container, along with local SQLite state and workspace files on infrastructure you control.
The public pages mention Codex, Claude Code, and Hermes as supported examples. The docs and guides also discuss how to run Claude Code and Hermes through an API.
The product is built for session-based agent work with streaming, file handling, artifacts, cancellation, and error handling. The examples page shows a Cursor-style coding workflow, and the Starter Kit includes other product patterns.
The pricing page says you can build and integrate for free, and that 500 launch credits are available for a limited time after adding a card. It also lists a Developer plan starting at $20 per month.
流量数据仅供参考。
aakar-ai.dev
在一个仪表板中管理 AI 供应商、项目配置、日志和分析,支持 BYOK 工作流。
hunyuan.tencent.com
Tencent Hy 是腾讯旗下的 Hy AI 产品、模型和工作室入口站点。
tongyi.aliyun.com
通义实验室是阿里云支持的 AI 模型与应用入口,集中展示 Qwen、Wan、最新动态及开发接入路径。
humanornot.ai
AI21 是一个面向企业的基础模型与 AI 系统平台,旨在帮助团队运行 AI 智能体和工作流,并兼顾质量、成本与可靠性。
ludwig.ai
开源声明式深度学习框架,通过单个 YAML 配置训练、微调和部署模型。
gemini.google.com
Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber 是面向开发者的 Google Gemini 模型,适用于构建生产级 AI 智能体,专注于效率、延迟、计算机使用和专业安全工作流。