Installation & system requirements
LYNX is in private beta (pre-release). Details may shift before general availability.
LYNX is one C inference core with a thin, idiomatic binding per language — the same models and the same open → predict → detections shape everywhere, from a Swift app to a Python script.
Install — Python
1pip install lynx --extra-index-url <your beta index URL>Python 3.10–3.14 · import lynx. The beta index URL comes with onboarding.
Supported platforms & versions
SDK version: 1.0. The build/patch number increments over time; confirm the exact build at runtime with version() (Lynx.version() / lynx.version() / lynx_version()).
CPU-only inference is supported on every platform via ONNX Runtime — no GPU required, at reduced throughput. Python wheels are selected automatically by pip. Jetson compiles to TensorRT on first run and caches the engine (JetPack 5 & 6).
Python
| Platform | Architecture | Min version | Acceleration |
|---|---|---|---|
| Linux | x86_64, aarch64 | Python 3.10–3.14 | CUDA / TensorRT · CPU |
| macOS | Apple Silicon, Intel | Python 3.10–3.14 | CoreML · CPU |
| Windows | x86_64 | Python 3.10–3.14 | CUDA / DirectML · CPU |
| Jetson | aarch64 | Python 3.10–3.14 · JetPack 5/6 | TensorRT |
Offline operation
Inference runs on-device — no per-frame network calls, no cloud dependency, and your video never leaves your hardware. Telemetry is optional and can be turned off. Trial verification uses periodic connectivity when available; paid licenses can be configured for extended offline operation. See the licensing page for how activation works.