FreeCode DeepSeek Harness

FreeCode DeepSeek Harness is a Windows x64 desktop application for using the DeepSeek Harness coding agent with a local project. It combines the DSH interface with an Electron shell and provides workspaces, coding sessions, tools, MCP server settings, and an embedded browser.

Why I built it

I wanted a straightforward Windows desktop workflow for coding with the Harness, without requiring users to install and maintain a separate gateway stack.

How it works

The current source version is 1.0.1. It bundles Phoenix, a local gateway built from pinned FreeLLMpool 0.13.0 source, and offers Auto Phoenix plus explicitly pinned free-model routes. The older OpenCode2API gateway was retired after upstream outages, proxy failures, rate limits, and breaking API changes. Phoenix does not require separately installed Python, C:\LocalAI, FreeLLMpool, or a pre-existing local service.

The repository's current catalog snapshot includes routes such as MiniMax M3.1 Flash, whose model-native context window is 1M tokens. This is not a promise that every provider route accepts the full window: upstream availability, rate limits, and effective context limits can change and may be lower. Internet access is required for provider calls. On a first launch without a valid cache, the app downloads and verifies the pinned DSH runtime; clean first-run use depends on that runtime asset being available.

What I learned

The main challenge was making gateway setup self-contained while keeping model selection predictable. Choosing a concrete model pins that route; the app does not silently switch providers. Only Auto may try another eligible route before streaming begins. The bundled gateway and the separate DSH runtime are verified before startup.

Platform

Windows x64 only. Windows ARM64, Linux, and macOS are currently untested and unsupported. This is a solo project maintained by an independent developer.

Built With

• ai
• deepseek
• electron
• mcp
• typescript
• windows