VSCodium-Rust
A local-first, agentic IDE with a Rust and Tauri backend and a React frontend, built around a local inference server tuned for AMD hardware, aimed at security researchers and cross-platform developers who want AI coding assistance without a cloud dependency.
VSCodium-Rust is a VS Code-shaped editor whose agent loop, code indexing and process management run in a native Rust process rather than an Electron main thread. Local inference runs through Lemonade, an OpenAI-compatible server running llama.cpp directly with no Ollama wrapper, tuned for AMD hardware through ROCm, with an optional split between a larger reasoning model and a smaller, faster operator model for sub-agents.
Optional services from a separate AGPL-3.0 submodule called kortex can sit in front of the backend: a KV-slot cache that skips re-prefilling repeated prompt prefixes, and a retrieval proxy that builds a dense catalog of the workspace and prepends only the most relevant chunks to a request instead of the whole repository. The editor, agent loop, indexing and local inference client are MIT licensed; cloud AI routing, hosted compression and subscription management are separate hosted services not included in the source. The project states that architecture, technical approach and merge decisions are human, with AI-generated code reviewed and tested against real hardware before it lands.