Checks a claim against a source you can inspect (a document, a dataset, a report or the open web) and answers confirmed, contradicted or clouded when it can't tell, with the quote or recomputation behind the answer. Pick an example below, or upload your own file.
| Model on Nebius Token Factory | Used for |
|---|---|
| nvidia/Nemotron-3-Ultra-550b-a55b | NVIDIA open model. substrate: reads the datasheet or your document and proposes the figure. searchlight: judges whether a quote supports or contradicts the claim. |
| openai/gpt-oss-120b | searchlight: reads fetched pages and proposes the quote. |
| Qwen/Qwen3-Embedding-8B | Ranks passages by meaning, alongside BM25. |
quant and silicon make no model call: statistics and timing figures are computed from the file. Endpoint: api.tokenfactory.nebius.com/v1 (OpenAI-compatible).
Pick a real example claim — it fills the form below, it doesn't run it
Stored only in this browser, on this device.
Points at the deployed engine by default. Change to http://localhost:8000 for local testing.
Bundle several claims — each can point at a different uploaded document — and get one rolled-up answer