While you wait

Data-use extraction runs entirely in your browser on WebGPU — no server, no uploads. It spots data mentions in research text, then attributes each with provenance and usage/impact.

GLiNER — detects spans, labels specificity
LFM2.5-350M — provenance (producer, year, geography, acronym) + usage & impact

Private by design. Nothing is sent anywhere — your text never leaves your browser.

DATA USE EXTRACTION

WebGPU · GLiNER + LFM2.5 multitask

Initialising…

GLiNER

span

huggingface/rafmacalaba/gliner_datause_extended-onnx

onnx/gliner_datause_extended_fp16.onnx

LFM2.5-350M

attribution

huggingface/rafmacalaba/lfm2.5-350M-datause-multitask-onnx

onnx/lfm2_multitask_fp16_v2.onnx

WebGPU pipeline 0%

Loading models…

Input text

stage 2 → 3

Paste research text, or load a sample — real-world research excerpts or synthetic examples.

EXAMPLES

Span threshold 0.60

WebGPU · client-side · no server

Extracted mentions

Spans as a tree — provenance and usage/impact. Toggle JSON for clean extraction or Trace for timings.

No mentions yet

Paste text or pick a sample, then press Extract.