CSGLite
A single-binary local LLM inference platform, inspired by Ollama, that downloads and runs models from the CSGHub platform with a web UI, an OpenAI-compatible API, resumable downloads and one-click coding-agent setup.
CSGLite runs large language models locally from a single cross-platform binary, downloading models from the CSGHub platform, hub.opencsg.com or a private deployment, and serving them through llama.cpp with GGUF models or auto-converted SafeTensors checkpoints. A single command downloads, loads and starts chatting with a model; loaded models stay warm for a configurable keep-alive period for instant reconnects, and a background API server starts automatically without manual setup.
Its web UI covers a model library with pause and resume downloads, a marketplace for browsing CSGHub models and datasets, an interactive chat interface, one-click configuration for AI applications, and settings for storage, third-party providers and access tokens. It can also form a LAN cluster of several machines sharing one secret, routing requests to whichever node finishes soonest and copying models between nodes instead of re-downloading them, and integrates with third-party providers such as OpenAI, DeepSeek and OpenRouter as well as coding agents including Claude Code, Codex, Pi and OpenCode.