Xyntetik Runner
Single-binary C GGUF engine that serves models and trains LoRA adapters locally.
Tested on macOS and Linux
Use it
macOS arm64
Download macOS arm64 binary from GitHub releases
curl -LO https://github.com/Joakimpalm-Zen/xyntetik-runner/releases/latest/download/runner-macos-arm64 && chmod +x runner-macos-arm64 && mv runner-macos-arm64 runner && ls -lh runnerDownload tiny GGUF model stories260K for testing
mkdir -p ~/models && curl -fL -o ~/models/stories260K.gguf https://huggingface.co/ggml-org/test-model-stories260K/resolve/main/stories260K-f32.gguf && ls -lh ~/models/stories260K.gguf
Check that it works
./runner --versionThe output should contain “runner 1.10.0”.
First thing to try
./runner --info capsRun ./runner --info caps to see what the engine supports on your machine.
Linux x86_64
download Linux binary and checksums with verification
curl -fL -o runner-linux-x86_64 https://github.com/Joakimpalm-Zen/xyntetik-runner/releases/latest/download/runner-linux-x86_64 && curl -fL -o SHA256SUMS https://github.com/Joakimpalm-Zen/xyntetik-runner/releases/latest/download/SHA256SUMS && sha256sum -c --ignore-missing SHA256SUMSmake executable, rename to runner
chmod +x runner-linux-x86_64 && mv runner-linux-x86_64 runner && ./runner --version
Check that it works
./runner --info caps 2>&1The output should contain “architectures”.
First thing to try
./runner --info caps 2>&1Run ./runner --info caps to see what the engine supports on your machine.
What it does
- Serve a model through an OpenAI-compatible APITested
How to use it
- Start the server with
./runner -m <model.gguf> --serve --port 8080. - Send a chat request with `curl -s http://127.0.0.1:8080/v1/chat/completions -H 'Content-Type: application/json' -d '{"messages":[{"role":"user","content":"<your text>"}],"temperature":0,"max_tokens":1
- Start the server with
- Train a LoRA adapter through quantized weightsNot tested
- Finish tool calls past the token limitNot tested
- Record a run as a signed receipt and replay itNot tested
How far pickfit got
macOS arm64
- Installs
- Starts
- 1 of 4 features work
- Works offline
Linux x86_64
- Installs
- Starts
- 3 of 4 features work
- Works offline
Other systems
Windows: the project says it runs there; pickfit hasn’t tried it.
Good for
- serve a GGUF model via OpenAI-compatible HTTP API
- check whether a model fits before downloading
- quantize model file with custom type plan
- train LoRA adapter through quantized weights
- verify signed reproducible inference receipt
An alternative to llama.cpp, vLLM, Ollama, TensorRT-LLM and SGLang.
How pickfit tested it
Each run starts from a clean machine and follows the steps above. How pickfit verifies
macOS arm64
It produces model output
- Replayed from scratch
- Works offline
- Checked on 1 Nov 2026
- Version 1.10.0
Differences the test machine hit
- README: Published release 1.6.1
In the test: Downloaded binary reports 1.10.0, still compatible
Linux x86_64
It produces model output
- Replayed from scratch
- Works offline
- Checked on 30 Oct 2026
- Version runner 1.8.0
Differences the test machine hit
- README: git clone + make; quick-start shows runner-macos-arm64
In the test: used MANUAL Linux asset runner-linux-x86_64 + SHA256SUMS with sha256sum -c; no compiler or make in box so source build impossible and macOS asset cannot run on Linux - README: ./runner --version -> runner 1.6.1
In the test: binary printed runner 1.8.0