Lenz

Product Introduction

1. Definition: Lenz is an audit-grade AI fact-checking API and software-as-a-service (SaaS) platform. Technically, it is a multi-stage, multi-model adversarial verification pipeline designed to ground AI outputs in real-world sources.
2. Core Value Proposition: Lenz exists to eliminate AI hallucination in production applications where accuracy is non-negotiable. Its primary value is providing verifiable, source-cited truth assessments for AI-generated or user-submitted content, moving beyond a single model's "best guess" to an auditable, consensus-driven verdict.

Main Features

1. /extract API Endpoint: This feature programmatically identifies and isolates verifiable factual claims from any block of unstructured text. It works by using specialized language models to parse input and output discrete, checkable statements, serving as the foundational step for automated fact-checking workflows. This is a core primitive for AI incident triage and content pre-screening.
2. /verify API Endpoint (Full Adversarial Pipeline): This is Lenz's flagship feature. It runs a submitted claim through an 8-model, 5-stage verification process. The technical workflow involves: claim decomposition, independent web search for source material, a structured multi-model debate where LLMs argue opposing sides, synthesis of arguments, and final review by three independent AI reviewers. It returns a scored verdict (e.g., 10/10) with a complete citation trail, showing every source and step in the reasoning chain.
3. /ask API Endpoint: This feature enables follow-up, source-grounded Q&A on any completed verification. It allows developers or end-users to ask contextual questions about the claim or verdict, with answers strictly constrained to the evidence and reasoning produced during the initial /verify process, preventing drift or new hallucinations.

Problems Solved

1. Pain Point: It directly addresses the critical issue of AI hallucination and unverifiable outputs in production systems. For applications in legal, financial, medical, or news domains, a single incorrect statement can lead to loss of trust, legal liability, or financial damage.
2. Target Audience: Primary users are AI product teams and developers building high-stakes applications (e.g., legal tech AI, financial analysts, customer support bots, content moderation systems). Secondary users include researchers, journalists, and educators who need to quickly verify claims with transparent sourcing.
3. Use Cases: Essential for AI incident triage (diagnosing errors in live outputs), asynchronous runtime verification (checking AI-generated content before user delivery), pre-release continuous integration (CI) checks for AI features, and powering a "fact-check" button within any knowledge-intensive application.

Unique Advantages

1. Differentiation: Unlike standard AI tools (e.g., ChatGPT, Perplexity AI) that provide a single model's answer from its training memory, Lenz employs an adversarial, multi-vendor LLM pipeline. It doesn't just retrieve; it debates and reviews, ensuring no single model's blind spots dictate the conclusion. Unlike manual fact-checking, it provides API-speed audits.
2. Key Innovation: The core innovation is its "audit-grade" methodology modeled on human rigorous review. The combination of multi-model debate (forcing opposing viewpoints) and a final review panel of three independent AI reviewers creates a system designed for consensus and error-checking, with every output linked to a full citation trail for human-in-the-loop verification.

Frequently Asked Questions (FAQ)

1. How does Lenz prevent AI hallucination in its own fact-checking process? Lenz mitigates self-hallucination through its adversarial pipeline design. By using multiple, competing LLMs from different vendors to debate evidence and requiring consensus from independent reviewer models, it cross-checks its own work. All conclusions must be grounded in the sources retrieved during the search stage, creating a verifiable audit trail.
2. What is the difference between Lenz's /assess and /verify API calls? The /assess endpoint is a faster (5-10 second), lighter-weight check using a 3-model consensus, ideal for synchronous user experiences where speed is critical. The /verify endpoint is the full, comprehensive audit, running the complete 8-model adversarial pipeline with source citation, taking 60-90 seconds for high-confidence, audit-grade results.
3. Can I use Lenz to fact-check live AI chat outputs or documents automatically? Yes, this is a primary use case. Developers can use the /extract API to pull claims from streaming or completed AI responses, then route them to the /assess or /verify APIs. This enables real-time or post-hoc fact-checking within chatbots, summarization tools, and content generation platforms.
4. Is Lenz available as a Model Context Protocol (MCP) server? Yes, Lenz offers an MCP server integration, allowing AI coding assistants and agent frameworks like Claude Desktop, Cursor, and others to directly access its fact-checking capabilities within their workflow, enabling developers to verify code-related claims or documentation.