Lenz is an audit-grade fact-checking platform designed to reduce hallucinations in AI products through an adversarial multi-model verification pipeline.
What the product does and how it is positioned
Lenz provides a suite of tools for AI product teams and individuals to verify the accuracy of generated content. By employing an adversarial pipeline that includes multiple models and independent reviewers, the platform aims to ground AI outputs in verifiable sources.
The platform offers both a direct interface for checking individual statements and an API for developers to integrate fact-checking capabilities into their own workflows. It supports various use cases, including incident triage and pre-release verification.
Source-supported ways to use the product
Teams use the platform to verify claims during incident response to ensure accuracy.
Developers integrate the API into continuous integration pipelines to perform automated runtime verification before software releases.
The documented workflow, where available
The system pulls verifiable claims from provided text using the /extract primitive.
The platform runs an adversarial pipeline involving multiple models and reviewers to assess the validity of the extracted claims.
Users perform follow-up questioning grounded in the verification results using the /ask primitive.
The core of the Lenz platform is its multi-stage verification process. Rather than relying on a single model, it employs eight different models across five distinct stages to evaluate the truthfulness of a claim.
To ensure high-quality results, the process incorporates opposing-side debate and utilizes three independent reviewers. This methodology is designed to produce a full citation trail, allowing users to trace the reasoning behind a verdict.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
Lenz is designed to provide audit-grade fact-checking for AI products, helping to detect and reduce hallucinations by grounding outputs in real sources.
Yes, the platform allows users to verify a single statement, link, or question by pasting it into the interface to receive a sourced verdict and reasoning.
Lenz is documented to run within Claude, Cursor, n8n, and Zapier, and it provides an API for custom integrations using OpenAPI 3.1.
The process uses an adversarial pipeline involving eight models across five stages, incorporating opposing-side debate and three independent reviewers to generate citations.
The API includes four primitives: /extract for pulling claims, /assess for multi-model verdicts, /verify for the full adversarial pipeline, and /ask for grounded follow-up Q&A.
