Runs local LLMs on an AMD Ryzen AI laptop (CPU, Radeon iGPU and XDNA2 NPU) behind one
OpenAI- and Anthropic-compatible API, with a desktop app to pick, install, try and
benchmark models. Design and status: docs/SPEC.md.
Download contio_<version>_amd64.AppImage from the
releases, then:
chmod +x contio_*_amd64.AppImage
./contio_*_amd64.AppImage # desktop app; starts the daemon if it isn't running
./contio_*_amd64.AppImage daemon # only the daemon (API on http://127.0.0.1:8080)
./contio_*_amd64.AppImage cli ps # the command-line clientThe app starts the daemon if none is running and stops it when you close the window.
To keep the API up without the window and start it at login, turn on Settings →
Background service → Run in the background: it installs contiod as a systemd user
service (binaries in ~/.local/share/contio/bin, refreshed when you start a newer
AppImage). Start/Stop are there too, and in the header when the daemon is offline.
Inference backends (llama.cpp, whisper.cpp, FastFlowLM) are set up from the app the first time
a model needs them; NPU support needs the XRT runtime (docs/spikes/m0-npu.md).
cargo build --release # contiod (daemon) and contio (CLI) in target/release/
cd apps/gui && npm install && npm run tauri dev # desktop app against a running contiodOptional config: ~/.config/contio/config.toml (docs/config.example.toml).
scripts/release.sh 0.4.0 # sets the version everywhere, commits, tags v0.4.0
git push origin main v0.4.0 # GitHub Actions builds the AppImage and publishes the releasedocs/PERFORMANCE.md is the performance roadmap: each item is measured on this hardware before it ships
as a setting under Expert → Performance in the model editor (or contio set <model> --spec ngram,mtp).
Licensed under either of Apache License 2.0 or MIT, at your option. Unless you state otherwise, any contribution you submit for inclusion in contio is dual licensed as above, without additional terms.
Benchmark data from GSM8K, HumanEval, MMLU-Pro and LibriSpeech keeps its own license; see THIRD_PARTY.md.