Janus Data sells whole-company corpora recovered from U.S. business liquidations. Design documents, source code, CAD, and correspondence that never reached the public web, licensed cleanly and delivered ready to ingest.
When a business liquidates, its records become a sellable asset. We acquire the complete corpus: everything the company wrote, drew, built, and argued about over its life. None of it was ever public.
Specifications, RFCs, architecture reviews, and test plans, with the revision history that shows how each decision was reached.
Native CAD assemblies, blueprints, and engineering drawings, paired with the change orders and review notes that describe them.
Private repositories with full commit history, code review threads, issue trackers, and CI configuration. Code that has never appeared in any scrape of GitHub.
Product roadmaps, project plans, budgets, and postmortems. Planning documents next to the outcomes they predicted.
Email and messaging archives that connect the documents to the people and decisions behind them, de-identified before delivery.
Financial models, manufacturing travelers, support tickets, and process documentation. The texture of a real business at work.
These documents lived on private servers and left them only through a court-supervised sale. They appear in no crawl, no public dataset, and no competitor's training run. Each corpus expands the frontier of what any model has read.
We sell corpora, not shredded documents. The spec references the drawing, the code implements the spec, the emails argue about both. Directory structure, cross-references, revision history, and authorship metadata all survive intact.
Lots sell in limited counts, and single-buyer terms are available. Novelty is the product, so licensing is structured to protect it. What you train on, others do not.
Every corpus ships deduplicated and indexed, with a manifest covering provenance, file types, token counts, and de-identification treatment. Text is extracted to structured formats with layout and metadata preserved; original binaries ride alongside.
Further processing is available to spec. Tell us the target and we deliver to it.
Executable environments built from real codebases, issue histories, and internal workflows.
Prompt and response pairs derived from tickets, reviews, and correspondence.
Paired drawings, CAD renders, and their written descriptions and change orders.
Databases and trackers exported to clean relational or document formats.
Labs walk away from data they cannot get honest reps on. We built the pipeline so you never face that choice. Every lot arrives with the legal work already done and documented.
We work with a network of Chapter 7 bankruptcy trustees across the United States. Their statutory duty is to convert estate assets to money for creditors, and a company's data is an estate asset. We prepare that asset for market and bring it to you. The flow is continuous, and the supply never repeats.
Tell us what you are training and we will match you to current and upcoming lots. Sample corpora are available under NDA, formatted the way you need them.
Email inquiries@janusdata.ai
Correspondence P.O. Box 60417, Palo Alto, CA 94306