New York, NY — July 24, 2026
New York, NY — July 24, 2026 — Xorq Labs today announced that its verification engine for AI working on enterprise data enters private preview on September 1, 2026. The first way in is Xorq Desktop, a native client that lets a data team meet the engine on their own machine. The same engine also runs headless in their own cloud — so what a practitioner tries locally is the same verification a company deploys at scale.
Xorq checks the numbers an AI produces against a team’s own data, so a hallucinated value is caught and corrected before it reaches production — and every answer carries traceable lineage back to the sources the team already trusts. AI is now fluent enough to state a revenue figure or a risk exposure with total confidence. Whether that number is correct is a separate question — and in shift-left, regulated work, even a correct number no one can trace is a liability, not an insight.
“The engine is the product, and it runs wherever the work happens,” said Hussain Sultan, Founder & CEO of Xorq Labs. “Xorq Desktop is the easiest way to see it work: point it at your data, ask a question, and get an answer with a certificate and lineage behind it. The same verification then deploys headless in your own cloud, unchanged. The desktop is optional. The verification is not.”
Xorq works with any model — open source, proprietary, or one from a major provider — and across any data engine (e.g. Snowflake, Databricks, +20), all on a team’s own warehouse, with no data migration:
Underpinning it all is Xorq Memory, an open source executable semantic catalog: every expression stores its schema, SQL, and sources and is content-addressed, so agents compose new work only on expressions a human has blessed and reuse cached results instead of starting over. Results are cached and reused, so the warehouse never bills the same answer twice.
On DABStep — a public benchmark of 450 questions over payment data — a smaller model paired with Xorq beats a larger model working alone. Anthropic’s Claude Haiku with Xorq Memory scores 84%, ahead of the 75% Claude Sonnet baseline, using the same model and prompt; only the verified work changes. Xorq also cut token usage by roughly 60%. Better answers at lower model spend.
Luke Pendergrass, Product Lead at Xorq Labs, will speak on making AI answers trustworthy enough to ship in the Startups & VCs track at Ai4 2026 in Las Vegas, August 2–5.
Xorq Memory is open source, so teams can trace how every number is produced instead of trusting a black box. Xorq enters private preview on September 1, 2026 — first through Xorq Desktop for waitlist members and early subscribers, with headless enterprise deployments run alongside the Xorq team. Join the waitlist at xorq.dev, or explore the code at github.com/xorq-labs/xorq.
Xorq Labs builds the verification engine for AI on enterprise data. Xorq checks the numbers an AI produces against a team’s own data, provides traceable lineage back to trusted sources, and caches results so the same work never runs twice — all on the team’s own keys and warehouse, with an open-source core. Xorq is backed by Lunar, Encoded, and Composed. Learn more at xorq.dev.
Copy our address and paste it into whichever email client you use.