QWED is an open-source verification infrastructure designed to help organizations check AI outputs before they reach production. Positioned between AI models or agents and operational systems, it independently re-derives claims, reviews tool calls, and assesses code, schemas, and agent state. Its fail-closed approach allows only proven results to proceed; outcomes are marked VERIFIED, UNVERIFIABLE, or BLOCKED.

The platform offers a suite of specialized verification engines covering mathematics and finance, formal logic, reasoning, statistics, factual claims, knowledge graphs, code security, SQL safety, structured schemas, and image metadata. It uses established tools and methods such as SymPy, Z3, SQLGlot, CrossHair, JSON Schema, and pandas with Pandera. Additional capabilities include taint analysis, policy rule enforcement, and checks for tool calls and agent workflows.

QWED provides documentation, open-source distribution, and a Docker container, alongside security and verification integrations within the GitHub ecosystem. Its stated purpose is to provide evidence-backed checks that complement approaches such as retrieval, guardrails, fine-tuning, and prompt engineering, helping teams assess AI decisions before they are acted upon.