nulity.ai
…
The product line

One foundation.
Every IP registry.

The world's most valuable text and image registries are public — the USPTO patent corpus, every issued trademark, every design figure ever published. We turn them into search infrastructure that knows what legally matters, not just what looks similar.

Why this is hard

Keyword search doesn't know which prior art actually wins an IPR. Generic embedding search doesn't know that a coffee logo near a tech logo isn't a conflict. The hard part of IP search isn't retrieval — it's the calibration layer that knows what legal similarity looks like. That's the layer we build.

Private betaPatent invalidation

Nulity Search

The references that invalidate, not just the ones that match.

Drop in any US patent. The system surfaces the prior-art candidates most likely to invalidate its claims — ranked by an outcome-aligned model and explained passage by passage. Built for IPR petitioners, defense counsel preparing patent-owner responses, and patent acquisition due diligence.

What's inside

  • Outcome-aligned rankerBi-encoder fine-tuned on PTAB Final Written Decisions, blended with shared-CPC and shared-citation signal. +44.5% recall@10 over BM25.
  • Matters workspaceGroup every search, saved reference, and note under a named case. Triage references shortlist / keep / dismissed; carry context across sessions.
  • Inline reading modeRead full patents inside Nulity — abstract, claims, description — without leaving the workspace. Out-of-corpus pubs fetch on demand.
  • Highlight + save passagesSelect any sentence in the reader; pick a color; it persists across sessions and shows up under the matter for claim-chart drafting.
  • Side-by-side compareDiff two references — claims, description, shared CPC codes, shared citations. Foundation for §103 obviousness combinations.
  • Forward citationsSee which other patents in the corpus cite this reference. The credibility signal that distinguishes a load-bearing prior art from an obscure miss.
  • Word-doc exportOne click pulls every result + structural evidence + top passages into a .docx that drops straight into an invalidity contention draft.
Request access
ResearchTrademark conflict

Nulity Brand

See trademark conflict in your creative before legal does.

Upload a campaign image, video frame, or product packaging. The system surfaces the closest registered marks, weighted by genuine infringement risk — not raw visual similarity. Designed for in-house brand teams and ad platforms shipping hundreds of creatives a week without a paralegal in the loop.
ResearchDesign + visual prior art

Nulity Visual

The first prior-art search that actually sees.

Design patents and figure-heavy mechanical claims have always been the blind spot of text search. A domain-tuned vision model turns “has anyone shipped something that looks like this?” into a single-image query.
The thesis

Public registries are infrastructure.

Patents and trademarks are the most legally consequential public data on earth. They've been frozen behind keyword search for forty years because building search that understands them requires a model that's read the entire corpus, a corpus big enough to train on, and a calibration layer aligned with actual legal outcomes. We're building all three.

See how the model works