Single-file local LLM runner. No backends, no agents, no GUI — just pip install and go.
pip install -e .
capybara pull bartowski/Meta-Llama-3-8B-Instruct-GGUF
capybara run llama3 "hi"Python 3.9+
git clone https://github.com/Distendo/capybara.git
cd capybara
pip install -e .Requires a llama-server binary on PATH or in ./bin/.
| Command | Description |
|---|---|
capybara pull <model> |
download from HF (or ollama/name:tag) |
capybara run <model> [prompt] |
chat interactively or one-shot |
capybara serve [--model M] [-F] |
start API + web UI (-F = foreground) |
capybara ui |
open Web UI in browser |
capybara list / ls |
list installed models |
capybara show <model> |
model details |
capybara rm <model> |
remove a model |
capybara cp <src> <dst> |
copy model |
capybara create -f Modelfile <name> |
create from Modelfile |
capybara search <query> |
search HuggingFace for GGUF models |
capybara ps |
server status |
capybara stop |
stop server |
capybara logs [-n N] |
tail engine log |
Running an uninstalled model pulls it automatically.
Pass native llama-server flags after --:
capybara run smollm -- --n-predict 512 --temp 0.7 --threads 8- OpenAI-compatible
POST /v1/chat/completions - Ollama-compatible
POST /api/chat,POST /api/generate,GET /api/tags - Built-in chat UI on the same port
python -m build
twine upload dist/*No Python dependencies beyond the standard library.
Apache-2.0