Developer Cactus Compute has released Needle 2, a lightweight AI model optimized for the Raspberry Pi 5 processor. The model file size is 14 MB and it does not require additional AI acceleration hardware to function.
Needle 2 is designed for local device control rather than general conversation. In tests on an 8 GB RAM Raspberry Pi 5, the model consumes approximately 28 MB of memory, with a total footprint of 46.4 MB when running a full Python demo.
The system translates natural language into function calls. Executing a command to turn on an LED takes 78 milliseconds, while querying the processor temperature takes 149 milliseconds. The model operates entirely offline without cloud APIs.
The model uses a Python annotation system to map natural language to specific hardware functions. If a user provides an input that does not match a predefined function, the system returns a null value within 92 milliseconds.
Needle 2 is available on Hugging Face, and its source code is published on GitHub under the Apache 2.0 license.
