heim¶

Build a full local library of LLM models, then run, chat, and serve them — entirely on your own hardware. Discover models from Hugging Face, Ollama, and LM Studio, pull them into one library on a drive of your choosing, and serve any of them through an OpenAI-compatible API and a browser chat UI.

heim web UI

flowchart LR
    hf["HF cache"] -->|heim library pull| lib
    ol["Ollama"] -->|heim library pull| lib
    ls["LM Studio"] -->|heim library pull| lib
    lib["Library on your drive<br/>gguf/ · mlx/ · cards + metadata"] -->|chat / serve --ui| rt["llama-server / mlx_lm.server<br/>OpenAI /v1 "]

Hold "Alt" / "Option" to enable pan & zoom

Proprietary, source-available

heim is not open-source. Copyright (c) 2026 Morten Hansen, all rights reserved (see LICENSE). The source is published for reference and evaluation; running it requires a written license — contact morten@winterop.com. Install is from source with uv (there is no pip install heim); see Getting started.

Why¶

One library, everywhere. A single, format-organized library on an external or cloud drive — move it by changing one setting.
Run anything, uniformly. GGUF via llama.cpp, MLX via mlx_lm — both behind one OpenAI-compatible endpoint, so any client attaches the same way.
Local and private. Nothing leaves the box. Built to pair with tool/MCP use by small local models.

Quick taste¶

uv sync
heim library ls                       # your library (the models on your drive)
heim library sources                    # models in app caches you could pull
heim library pull lmstudio <name>       # pull one into the library (--move to relocate)
heim serve --ui                 # browse + chat in the browser
heim chat <name> -p "hello"     # one-shot, scriptable answer
make run MODEL=<name>           # backend + browser UI, locked to one model

Start at Getting started, or jump to the CLI reference.

The name¶

heim is short, easy to type as a command, and deliberately neutral — it collides with nothing (no PyPI package, no product, no brand), which is the safest kind of name. If you want a meaning, read it as 鼓動 (kodō) — Japanese for heartbeat / pulse: the steady pulse of your own models, running on your own hardware.