Local intelligence. No cloud dependency.
A compact GGUF-quantized model built for edge hardware, regional language support, and work that keeps going when the network does not.
1.8 GB GGUF quantization
38 ms on consumer ARM hardware
32 African languages and regional dialects
None for local inference
Figures are supplied in Nion’s current project brief. Inference latency depends on the device and runtime configuration.
Put useful intelligence close to the work.
nil-3b is intended for local execution: no API round trip is needed for on-device inference, and the model can keep working when power or bandwidth is unreliable.
The target is practical computing on accessible hardware, shaped around regional languages and offline applications.
The model stays on the edge.
NIL-3B is designed to run from local memory on consumer ARM hardware. The local-first approach avoids a cloud round trip and keeps inference available without a connection.
Explore NIL Labs.
Visit the Nion developer gateway for available tools and access information.