coze
is a small egui
application for prompting a local offline LLM using the Huggingface candle
crate.
Currently it supports the following quantized models:
The current version supports:
- Prompt history navigation with fuzzy matching.
- History persistence across runs.
- Token generation modes.
- Copy prompts and replies to clipboard.
- Light/Dark mode.
See the app Edit/Config
menu for usage details.
The latest version of coze
can be installed or updated with cargo install
:
cargo install --locked coze
or by getting the binaries generated by the release Github action for Linux, macOS, and Windows in the releases page.
The first time it runs it will download the model weights from Huggingface to the ~/.cache/coze
folder.
To build locally (debug build may be very slow):
git clone https://github.com/vincev/coze
cd coze
cargo r --release