DeepSeek Harness developer preview
Points and comments are a snapshot, not live.
DeepSeek Harness developer preview offers a modular, plugin-based agent framework.
DeepSeek Harness is an open-source developer preview for building agent harnesses, where every capability (models, tools, skills, sessions, sandboxes, storage, loops, scheduling, UI) is a plugin built on the Cordis kernel. It features an append-only session log for traceability, with trajectory view, resume, fork, search, and replay. Multiple runtime modes are available: Standard, Code, Minimal (two-tool coding agent), and Creator mode for custom presets. The project is MIT licensed, source available on GitHub, and installable via npx @deepseek-ai/dsh web or from source.
What commenters are saying
Commenters questioned the choice of Node.js for agent harnesses, citing Python's fragility and distribution issues, Java's image problem, and Rust's lack of adoption. Supporters argued Node.js benefits from async design, cross-platform reach, fast iteration, and npm distribution ease. One commenter noted that Codex (OpenAI) is written in Rust, while others pointed to several Rust alternatives, including CodeWhale, Grok Build, Pi, and the multi-language Smol project. A counterpoint: if LLMs are writing the code, a compiled safe language might be better.