Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots
Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.
A 45M-parameter tool-calling model quantized to 2-bit fits in a 14MB binary and runs a full session in 28MB RAM, hitting 300–700 tok/s on sub-$200 phones and even ESP32-S3 microcontrollers — no GPU/NPU needed. It's purpose-built for function calling and structured extraction (not chat), enforcing output via a byte-level grammar compiled from your schemas and emitting a confidence score plus empty-call refusals so you can locally handle routine device control and only escalate ambiguous requests to the cloud. If you're shipping on-device agents for wearables, robots, or budget hardware, this makes private, offline tool routing viable where you'd previously have needed a cloud round-trip.
A 45M-parameter agentic LLM has been compressed to a 14MB binary, enabling it to run on devices with as little as 28MB of RAM, such as sub-$200 phones and microcontrollers, at speeds of up to 500 tokens/sec, allowing for on-device AI capabilities like tool calling and structured extraction without relying on cloud processing.