Needle 2 on a Mac
45M on-device model for tool calling and structured extraction · AI / LLMs
Needle 2 (Cactus Compute, Apache-2.0) is a 45M-parameter foundation model small enough to run on any Mac; the weights are a single 14 MB .cact file baked into the inference engine. It is built for tool calling, device use and structured extraction: text in, JSON out, with confidence-gated responses, built-in tool retrieval from large catalogs, and bounded memory from a 256-token sliding window. Install it with pip install cactus-needle; needle playground serves its own web UI on 127.0.0.1:7860. FrontierStack's Needle pane installs it, runs the playground and extracts structured fields from a passage of text. It also fine-tunes: needle generate-data, needle finetune and needle build produce a tuned .cact you can point the playground at (add the [train,metal] extra for Apple Silicon GPU training). Note it is not an OpenAI-compatible server and not an assistant model; it appears in the AI Models picker but is deliberately refused for the assistant's tools, because at 45M parameters with a 256-token window it cannot drive them reliably.
Run Needle 2 with FrontierStack
Install or connect Needle 2; then check its status, ports and certificate from FrontierStack. The same screen links to firewall checks, Malware Audit and backups where they apply.
Run it from your Mac.
FrontierStack installs, monitors and secures services on this Mac and on linked servers.
Download FrontierStack