Needle: a 14 MB model that turns commands into actions on a Raspberry Pi
When people talk about LLMs they immediately think of chatbots, but Needle takes a different path. It is a small AI model, just 14 MB, developed by Cactus Compute, with one job: turning a natural-language request into a call to a Python function. It is not built for conversation. It is built to control applications, devices and automations.
What caught my attention is that it runs locally on a Raspberry Pi 5 using only the CPU, with no cloud and no dedicated AI hardware. Instead of writing accendi_led(), you just say "Turn on LED 1" and the model picks the right function. It also works with Italian commands, and the project's website has online demos (home automation, robot vacuum, smartwatch, AR glasses) you can try.
The RaspberryItaly tutorial builds a small local assistant that saves notes to a file, reads the Raspberry Pi's temperature, switches an LED on and off, and understands commands written in Italian. You need a Raspberry Pi (preferably a 5), an up-to-date Raspberry Pi OS, Python 3, an LED with a 220 or 330 ohm resistor and a few jumper wires. The installation goes through a virtual environment and the cactus-needle package; the internet connection is only needed for the first download.
One tip from the author is worth remembering: it is better to give one command at a time. A sentence like "turn off the kitchen light and lower the blind" may not be interpreted correctly, while the two commands given separately work.
As a project idea I find it very practical: combined with local speech recognition, it would let you command a robot or a small automation without depending on any network service.
Source: RaspberryItaly
What caught my attention is that it runs locally on a Raspberry Pi 5 using only the CPU, with no cloud and no dedicated AI hardware. Instead of writing accendi_led(), you just say "Turn on LED 1" and the model picks the right function. It also works with Italian commands, and the project's website has online demos (home automation, robot vacuum, smartwatch, AR glasses) you can try.
The RaspberryItaly tutorial builds a small local assistant that saves notes to a file, reads the Raspberry Pi's temperature, switches an LED on and off, and understands commands written in Italian. You need a Raspberry Pi (preferably a 5), an up-to-date Raspberry Pi OS, Python 3, an LED with a 220 or 330 ohm resistor and a few jumper wires. The installation goes through a virtual environment and the cactus-needle package; the internet connection is only needed for the first download.
One tip from the author is worth remembering: it is better to give one command at a time. A sentence like "turn off the kitchen light and lower the blind" may not be interpreted correctly, while the two commands given separately work.
As a project idea I find it very practical: combined with local speech recognition, it would let you command a robot or a small automation without depending on any network service.
Source: RaspberryItaly
Comments 0
Leave the first comment, in any language.
Log in to comment