Data Truffle represents established companies licensing their operational records. We are seller-side and paid by the seller, so what you get from us is not a lead list. It is a manifest that has already been scored, priced against current demand, and cleared on rights, from a company that has decided to license.
We do not name buyers on this site. We name you to the seller, before any sample moves.
Described by category because that is what matters to a seller. The names stay between you and the company you are buying from.
Training general-purpose foundation models. Buy direct and through data labs. The most demanding on provenance and rights, and the most interested in breadth across industries.
Building training and evaluation sets under contract to several labs at once. Frequent buyers; often the first to be short on a specific modality or sector.
Fine-tuning for a vertical: accounting agents, support automation, legal drafting. Precise about industry and workflow; willing to pay for exclusivity in their field.
Building internal agents at large companies, often from adjacent industries or suppliers. Newer to licensing; strict on contract terms because their own counsel is.
Every manifest is scored on the same seven dimensions. The anonymized profile you receive shows the scores and the reasoning, so the first conversation is about fit, not discovery.
Which company, which systems, which years, produced by whom in the course of what work. Provenance decides whether the set is real and representative before anything else is looked at.
Can the seller license it? Employee and customer notices, client confidentiality, software terms. Rights are checked on the manifest, confirmed in diligence, and warranted in the license.
Years of the same workflow, with continuity. Eight years of one support desk beats two years each of four different tools.
How much of the real work left a trace: tickets, approvals, handoffs, call logs. Dense sets show the whole trajectory, not just the finished document.
Resolved or escalated, won or lost, paid or disputed. Outcome-labelled records train and evaluate judgment, not just language.
Counts per record type, with the distribution over time. Enough to matter, not so much that quality is unknowable.
Text, structured fields, audio, video, code. Most buyers are long on text and short on calls with outcomes and chat where problems get solved live.
Raw data, ever, from us. A sample moves from the seller to you under a mutual NDA once the seller approves your identity and intended use. Licensed records move from the seller's systems to yours after signing. We never hold a copy.
Sellers are told exactly this list, and told that you accepted it. It is the reason they are willing to license at all. A buyer who cannot accept it is not a buyer we can represent a seller to.
The form below: what you are training or evaluating, which industries and modalities you are short on, the volumes and years that matter, your appetite for exclusivity, and your timeline. We confirm the entity and the vetting standard, then add your demand to the signals we price against.
When a seller's manifest matches your demand, you receive the anonymized profile and scores: industry, size, years, systems, record types, volumes, how the systems link. No company name yet. You confirm or decline the indicative range.
The seller approves your identity and intended use. You sign a mutual NDA with the seller, receive a de-identified sample they chose, and run diligence. Same information and the same deadline as every other shortlisted buyer.
You submit binding terms. The seller compares offers on price and the six terms, chooses, and signs with you directly after counsel review. Records transfer from the seller's systems to yours. Our fee comes from the seller; you pay us nothing.
Specific beats broad. "Eight-plus years of resolved support tickets with call recordings, from US professional-services firms" gets matched; "operational data" does not.
We reply within two business days with a short call to confirm the entity and walk through the standard. Nothing you enter here is shared with a seller until a manifest matches and you have accepted the five terms.
Seven questions. A one-page manifest. A first offer from vetted buyers in 24 to 48 hours.
Measured from a complete data manifest. A first offer is a process target, not a guarantee; buyers decide eligibility, price, and terms.