Interactive demo of Spotlight Memory as implemented in spotlight-memory-synth-1b — 48.9M params trained on SYNTH.
Runs entirely in your browser: ~98 MB of fp16 weight shards download once, then inference executes through WGSL compute shaders on a WebGPU pipeline. No server; your prompts never leave the client. WebGPU is required (any current Chromium, Safari 18+, Firefox on supported platforms) — the app fails cleanly with guidance if the browser has none.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support