Show HN: CodexLocal – Offline, Privacy-First AI Coding Tutor (WebLLM and RAG) (CodexLocal.com)

CodexLocal is a new Show HN project that brings an offline, privacy-first AI coding tutor to your browser by combining WebLLM model execution with retrieval-augmented generation (RAG). Instead of sending your code or prompts to a cloud API, CodexLocal runs models locally via WebGPU in modern browsers (Chrome, Edge, Safari on supported OSes), using RAG to pull relevant snippets from your local codebase or docs and produce context-aware suggestions, explanations, and fixes. This approach matters for developers who need low-latency, private code assistance without exposing proprietary code to external services. Technically, it depends on WebLLM-style runtimes that leverage the GPU via WebGPU for inference; if your browser/device doesn’t support WebGPU you’ll see an error and must enable chrome://flags/#enable-unsafe-webgpu in Chrome/Edge, upgrade drivers/OS, or use a compatible browser. The trade-offs are clear: full offline privacy and reduced network dependency, but hardware and browser support are required and performance will vary by GPU. For teams who can meet those requirements, CodexLocal demonstrates a practical on-device workflow for RAG-enabled coding assistants.