AI training data catalog
The 24 types of data AI teams license most, what each is worth to buyers, where it comes from and what has to be cleared first. We source every type on request from vetted holders.
- This catalog describes categories of data we source and broker, not an inventory we hold.
- Each entry has a specification sheet: how it is measured, who buys it, where it comes from, the rights and privacy checks it needs, and published pricing signals.
- AI teams can request any type; data holders can list data of any type for a free assessment.
Language and text
Conversational dataConversational data is multi-turn dialogue between people, or between people and software, recorded with enough context to show how a conversation unfolds.Conversations, turns or tokensEmail archivesBusiness email archives show how real work gets done: requests, negotiations, approvals and follow-ups across long threads.Messages, threads, mailboxes and years coveredWorkplace chat dataWorkplace chat from Slack or Microsoft Teams shows fast, informal coordination that email does not capture.Messages, channels, active users and months of historyBooks and long-form textBooks and long-form text teach models to follow arguments and structure across thousands of words, which short web text cannot do.Titles, words or tokens, plus subject coverage and publication yearsMultilingual and low-resource language dataLow-resource languages have little high-quality text online, so licensed data in them is scarce and in demand.Words or tokens for text, aligned sentence pairs for parallel data, hours for speechBusiness documents and formsDocument AI models learn to read layouts, tables, signatures and fields from real business documents.Documents and pages, by document type and layout variety
Speech and audio
Speech and audio dataSpeech datasets train speech recognition, voice assistants and audio language models, and buyers value accent, language and acoustic variety.Hours of audio, number of speakers, languages and accentsCall center recordingsCall center recordings pair real customer problems with how skilled agents resolve them, which voice agents need to learn.Hours of audio and number of calls, by industry and language
Vision and video
Video training dataVideo model builders license footage they cannot scrape: unpublished, high-resolution and with clear rights.Minutes or hours, plus resolution, frame rate and subject categoriesImage datasetsImage datasets train image generation, vision and multimodal models, and captions describing each image raise their value.Images, plus resolution, subject coverage and caption qualityEgocentric and robotics videoRobotics and physical AI teams need first-person video of people doing real tasks with their hands.Hours of footage, task categories, number of distinct operators and environmentsMedical imaging dataMedical imaging datasets train diagnostic and triage models, and labels from qualified clinicians are what make them valuable.Studies, images or slides, by modality, body part and finding3D and CAD data3D and CAD data trains generative 3D, engineering design and spatial AI models, and there is very little of it online in usable form.Models or assemblies, part counts, categories and file formats
Code and agents
Source code datasetsPrivate codebases with real commit history, tests and reviews are worth more to coding model builders than public code, which most labs already have.Repositories, lines of code, commits, languages and years of historyAgent trajectory dataAgent trajectories are step-by-step recordings of a person completing a task in software: every click, keystroke, screen and result.Trajectories or tasks completed, applications covered, average steps per taskSoftware issue and ticket historyIssue trackers record how teams plan, triage and fix work, which coding and project agents need to learn.Issues, projects, years of history and percentage linked to code changes
Human feedback and evaluation
RLHF and preference dataPreference data records which of two or more model responses a person judged better, and why.Comparisons or rated prompts, by domain and rater expertiseEvaluation datasetsAn evaluation dataset tests a model on tasks it has never seen, with answers checked by qualified people.Tasks or items, with difficulty levels and domain coverageExpert-annotated dataExpert-annotated data is labeled or written by people with verified professional skill, such as clinicians, lawyers, engineers or accountants.Items labeled or expert hours delivered
Domain and enterprise
Customer support ticketsSupport tickets pair a real customer problem with the steps that solved it, which support agents and AI assistants learn from.Tickets, by product area, language and yearsCRM and sales dataCRM data shows how deals progress: activities, emails, calls, stage changes and outcomes.Opportunities or accounts, activities and years of historyFinancial documents and recordsFinance AI needs real statements, ledgers, reconciliations and filings to learn accounting judgment.Documents, entities and yearsLegal documents dataLegal AI models learn drafting and analysis from real contracts, briefs, memos and templates.Documents, by practice area, jurisdiction and document typeScientific and research dataScientific AI needs real experimental data, including failed experiments that never get published.Experiments, records or files, by field and method
Tell us what you need
Book a free 30-minute consultation. Tell us what data you want to buy or license, and we'll tell you honestly whether we can help and what it involves.
- We reply to every enquiry within two working days.
- No upfront fees for sellers
- We never hold your raw data