Needle 2 is an open 45M-parameter model for tool calling, device use, and structured extraction. The whole model is a single 14MB binary that runs a full session in about 28MB of RAM.
A 256-token sliding window keeps the tools pinned as KV sinks, so the number on the gauge stays flat however long the conversation runs. That is the property a wearable actually needs.
Weights are baked into the engine, there are no separate model files to manage, and inference does no network.
A byte-level grammar compiled from your schemas constrains every token, so tool calls come back as structured data.
Declare a large catalogue and a retrieval head renders only the top five tools per turn, with the grammar held to that subset.