Validiti · live concept demo
Search a movie. Or ask it a question.
You already know this story. Search it by keyword — always, with no AI at all — or, when a language model is available, ask in plain language and get a grounded, cited answer. Either way, the film is never opened to find the moment.
A 54-second storybook film — search it, or ask it a question below ↓
Search by keyword Always on · no AI needed
The baseline that never goes away — this runs on the substrate alone, with no language model and no network. Type a word and jump to the moment.
Ask in plain language Optional · needs a language model
When a language model is available, ask a question — it answers only from the moments the substrate retrieves: grounded, cited, never made up.
Grounded in these moments — tap to jump:
Two ways in — and one of them always works. Keyword search runs on the substrate alone: no AI, no network, so it keeps working offline, air-gapped, or anywhere a language model isn’t available. Add a model and you can also ask in plain language — and even then the film is never decoded: the model reads only the retrieved moments, and every claim cites a provable second. The reasoning is the model’s; the grounded, tamper-evident evidence is the substrate’s — sourced, not guessed. The same works across thousands of clips, with the video kept in the format you already use.