# LlamaStash

A terminal-native manager for local large language models that bundles a TUI, CLI, background daemon and OpenAI-compatible proxy around llama.cpp, with support for several other local inference backends.

LlamaStash is a single Rust binary that launches, supervises and routes local LLMs. It runs llama.cpp directly as its default backend instead of wrapping heavier tools such as Ollama or LM Studio, and it can also drive Lemonade for NPU or multi-engine inference, vLLM and SGLang for safetensors models, a dedicated backend for DeepSeek-V4, and any other OpenAI-compatible server declared in its configuration.

The same binary works as an interactive terminal UI, a scriptable CLI, and a background daemon exposing an OpenAI-compatible proxy endpoint, so other tools can talk to whichever local model is running. An init wizard installs llama-server for the current hardware, downloads a starter GGUF model, writes a tuned configuration, and smoke-launches it.

It targets Windows, macOS and Linux, and is documented for use both interactively and as a shell-tool surface for coding agents.

Website: https://llamastash.dev