A new open-source project called Kin-Code aims to replicate Claude Code's terminal-based AI coding experience in a single 9MB Go binary with no runtime dependencies.
Built by the team behind LocalKin (a self-evolving AI agent swarm), Kin-Code ships as one executable file that needs no Node.js, Python, or Rust toolchain. The project claims roughly 20MB of memory usage at runtime compared to what it estimates as 150MB for Claude Code.
The feature set covers the core Claude Code workflow: bash execution, file read/write/edit with diff visualization, glob and grep search, web fetch, and sub-agent spawning for parallel tasks. It supports three providers out of the box - Anthropic, OpenAI, and Ollama for running local models.
One notable feature: Kin-Code can authenticate through your browser using an existing Claude Free, Pro, or Max subscription, bypassing the need for a separate API key. It also includes MCP (Model Context Protocol) support, session persistence, and something called "soul files" for defining custom AI personas.
The project is MIT-licensed and very early stage - 18 commits, a single contributor, and no official releases yet. The comparison numbers against Claude Code focus entirely on binary size and memory footprint, with no coding benchmarks or accuracy comparisons. For anyone looking to experiment with a lighter-weight alternative or run AI coding assistance through local models, it's an interesting starting point. But calling it a Claude Code replacement today would be generous.