Product Workflows Customers Pricing Security Docs Sign in Book a demo

Litigo product · Verification

Citel

The independent citation-Verifier model for legal AI. Citel checks every citation, quote, and factual claim against primary law, flags unverifiable or overruled authority, and returns a signed verification report.

Book a demo → Read the API

Agentic verification Retrieval fact-checking Guardrails

Primary hardware · H100 / L4

The problem

Legal AI hallucinates citations, misquotes authority, and cites overturned cases. One bad cite in a filing risks sanctions and malpractice — yet no standalone, auditable verifier exists that firms can trust on any AI-drafted document.

What Citel checks.

Litigo's Verifier, productized: a dedicated model plus API that runs on Litigo output — and on documents from any other tool.

Verification

Citations, quotes, claims.

Every citation, quotation, and factual claim checked against primary law — not against a headnote or a paraphrase.

Negative treatment

Overruled detection.

Flags overruled, questioned, or otherwise negatively treated authority before it reaches a filing.

Report

Signed & timestamped.

A signed verification report drops straight into the audit trail.

API

Verification-as-a-service.

A metered API third-party legal-AI tools can call on their own output.

Confidence

Never silently dropped.

Confidence scoring; unverifiable claims are flagged, always surfaced to a human.

Built on the NVIDIA stack.

Layer	Technology	Why it's here
Hardware	H100 / DGX · L4 / RTX	Training the specialized Verifier model on citation-error datasets; high-throughput inference.
Libraries	CUDA · cuDNN · NeMo · NeMo Guardrails	Fine-tune the Verifier model and validate generative output.
Serving	TensorRT · Triton	Batch verification at low latency — fast enough for interactive drafting.
Platform	NIM · NVIDIA AI Enterprise	Deploy the verifier as a supported microservice, cloud or air-gapped.

Verifying a long document means cross-encoding thousands of claim/source pairs per filing under tight latency. GPU cross-encoders and TensorRT serving make per-filing verification fast enough to run inline; CPU-only cannot.

Why it holds up.

A proprietary error corpus

A growing corpus of citation errors and corrections compounds with every verified document — training a verifier competitors can't replicate from public data.

Trained on the failure modes

A model built specifically for fabricated cites, misquotes, and stale authority — the exact failures that erode trust in legal AI.

Neutral infrastructure

Sold as a neutral citation-integrity layer other vendors also buy. Malpractice carriers can consume the reports as audit evidence.

Metered, per claim

Attached to every Litigo document, then opened as an industry trust layer with usage-based API revenue.

Part of the ecosystem

Citel verifies the output of every product: Verdex rationales, Discovo privilege calls, Deposa digests, and Litigo drafts — all against the shared primary-law corpus.

Verify before you file.

See Citel check a real brief, live, in 30 minutes.

Book a demo →