needle
by cactus-compute
Needle is a very small AI model, between 8 and 29 MB, that turns plain-language requests into function calls and pulls structured fields out of messy text. It also produces text embeddings and, with its companion Whistle model, speech-to-text, all running locally on phones, devices and microcontrollers.
- Stars
- 14.4k
- Licence
- Apache-2.0
- Runs on
- Your computer
- Setup
- Moderate
- Language
- Python
- Age
- 246 days
What you can do with it
- Let users control an app or device by typing or speaking, with the model choosing which of your functions to call and filling in the arguments
- Extract typed fields such as invoice or booking details from messy text, with output guaranteed to parse
- Transcribe speech offline in seven languages and fine-tune the speech model on your own audio and vocabulary
Make money with it
A freelance developer could build and sell offline voice-control or form-extraction features for smart-home, kiosk or mobile app clients.
An idea, not financial advice. Check the licence before charging for anything built on it.
Ways a business could use it
- Add offline voice commands or natural-language controls to a product without paying per-request cloud API fees
- Automatically pull fields from invoices, bookings or notifications into your own systems, keeping customer data on the device
Can you use it commercially?
Free to use, change and sell, including in paid products. Keep the copyright notice.
Not legal advice. Read the project's licence file before building a product on it.
Getting it running
Install the Python package with "pip install cactus-needle", then decorate your functions and run the agent; a community npm package covers Node.js and browsers.
Follow the official instructions in the project README.
Things to check
- It is not a general chat model, as general chat ability is traded away for tool calling and extraction
- Whistle can turn music or other non-speech sounds into text, so an always-on microphone needs a voice activity detector
- Whisper base scores better on some speech benchmarks (TED-LIUM, AMI, MLS)
- Needle 3 is a new architecture, so check how well it handles your own tools before depending on it
Who's talking about it
Checked 28 Oct 2026. Summary drafted by AI from the project's README.
