--- title: "Dataset Licensing & Sourcing | Rights-Cleared Data | YPAI" url: https://ypai.ai/dataset-licensing/ description: "License rights-cleared datasets YPAI owns or represents, source what you cannot find, collect only the gap. Every record carries its consent and licence." source: "src/copy/routes (route /dataset-licensing/)" --- # The data may already exist. What you licence is the record that lets you use it. > License rights-cleared datasets YPAI owns or represents, source what you cannot find, collect only the gap. Every record carries its consent and licence. Rights-cleared inventory YPAI owns or represents, curated dataset sourcing through partners, and collection of only the gap. You describe what you need. YPAI finds what already exists in its own and partner inventory, shows what can be licensed quickly under NDA, and collects the rest to order. Every record arrives with its consent and licence basis attached, so your counsel can read where the rights come from before you sign. Example. A delivered record carries your agreement, your scope and your term. - Dataset Licence Agreement · current version - derived transcripts · models by written consent - fixed · stated in the schedule ## Four ways to buy. Every one begins with what already exists. The line below runs from data YPAI already holds to data that does not exist yet. Each way to buy sits where the data sits on that line, and the price of the work follows it. ### Off-the-shelf licensing - YPAI licenses data it controls with sufficient rights. You see the dataset card and a controlled sample under NDA, then the licence sets scope, term and territory. ### Partner-sourced licensing - YPAI builds a partner inventory and represents datasets from other suppliers. YPAI checks rights and availability, then packages and reviews the dataset before you see a sample. ### Bespoke sourcing - exists elsewhere · sourced to order - You name the data. YPAI identifies it, samples and technically reviews it, clears the rights, and arranges the licence and the delivery. ### Custom gap-fill collection - does not exist yet · collected - YPAI collects what the inventory does not cover: missing languages, accents, age groups, scene or session types, formats, annotation, rights scope. You license what exists and collect only what is missing. ## Find what exists. Collect only the gap. Below, one requirement moves through the flow. The matches light in the inventory, the gap is drawn around what is missing, and the trace from every record to its consent and licence basis stays attached the whole way. ### Find existing rights-cleared inventory - Own and partner inventory is searched against your requirement. ### Present what can be licensed quickly - Dataset cards, provenance descriptions and controlled samples, under NDA. ### Identify the gap - Missing languages, volumes or modalities are named against the requirement. ### Collect only the gap - The missing part is collected to order, under the same consent and rights controls. ### Keep the trace - Every record stays traceable to its consent record and its licence basis. The lit squares are matches in own and partner inventory. The frame is the gap, and only the gap is collected; every record keeps its consent and licence basis attached. ## Your counsel asks the questions. The record answers them. A licensed record answers which agreement authorised the data, which rights applied when it was made, and which uses are allowed now. Let the model read the example schedule below and it answers each question in this tab; you accept each answer. Data Licence Schedule, example. Licensed dataset: conversational speech recorded in home and vehicle settings, with transcripts and speaker metadata. Licensor authority: the dataset is licensed by YPAI, the Licensor, under contributor agreements recorded per contributor, with consent documented for each recording and withdrawable at any time. Authorising agreement: this Data Licence Schedule under the Master Services Agreement and its Statement of Work. Permitted use: training and evaluation of automatic speech recognition models by the Licensee. Derivative works: derived transcripts and acoustic features are permitted; derivative models offered to other parties require the Licensor's written consent. Sublicensing: requires the Licensor's written consent. Territory: processing and storage within the European Economic Area. Term: a fixed term stated in the order form, renewable by agreement. Exclusivity: non-exclusive. Evaluation samples: samples provided before signature are deleted at the Licensor's request once evaluation ends. Withdrawal: a contributor's withdrawal is propagated to the Licensee's copy under the remediation procedure. - Conversational speech recorded in home and vehicle settings, with transcripts and speaker metadata. - The dataset is licensed by YPAI, the Licensor, under contributor agreements recorded per contributor, with consent documented for each recording and withdrawable at any time. - This Data Licence Schedule under the Master Services Agreement and its Statement of Work. - Training and evaluation of automatic speech recognition models by the Licensee. - Derived transcripts and acoustic features are permitted; derivative models offered to other parties require the Licensor's written consent. - Sublicensing requires the Licensor's written consent. Processing and storage within the European Economic Area. A fixed term stated in the order form, renewable by agreement. Non-exclusive. - Samples provided before signature are deleted at the Licensor's request once evaluation ends. A contributor's withdrawal is propagated to the Licensee's copy under the remediation procedure. ### Which agreement authorises this use? - this Data Licence Schedule under the Master Services Agreement and its Statement of Work ### How is consent documented? - consent documented for each recording and withdrawable at any time ### Which use is permitted? - training and evaluation of automatic speech recognition models by the Licensee ### Where may the data be processed? - processing and storage within the European Economic Area ### May it be sublicensed? ### What happens to evaluation samples? - samples provided before signature are deleted at the Licensor's request once evaluation ends Let the model read the schedule extractive model · runs in this tab · a first pass, you decide A first pass by a model in this tab. You decide. - a dataset card and a datasheet - machine-readable provenance and permitted use, in the open dataset metadata format - the agreement set, versioned, with the consent version that was active - an integrity check per file and the withdrawal path - [the EU AI Act's data-governance duties for training data](https://artificialintelligenceact.eu/article/10/) - [the training-content summary general-purpose model providers must publish](https://artificialintelligenceact.eu/article/53/) - [the Commission's template for that summary](https://digital-strategy.ec.europa.eu/en/library/explanatory-notice-and-template-public-summary-training-content-general-purpose-ai-models) - [the Munich Regional Court ruling on memorisation in model training](https://www.justiz.bayern.de/gerichte-und-behoerden/landgericht/muenchen-1/presse/2025/11.php) - [Croissant, the dataset metadata format with provenance and permitted use](https://mlcommons.org/2026/02/croissant-1-1/) - [ODRL, the rights model the platform builds on](https://www.w3.org/TR/odrl-model/) ## Under NDA you see the inventory. The licence grants the use. The NDA is the first gate. It opens the inventory to you, and the licence that follows grants the use. Below, what each one covers. - partner and sourcing information where the agreement allows it YPAI can require deletion of samples and evaluation material. A Dataset Licence Agreement, or a Master Services Agreement with a Statement of Work and a Data Licence Schedule. - a fixed term or a renewing one - permitted use, training, evaluation, derivative models ## Speech, audio, image, video, text, sensor and multimodal. Each specialist page has the detail. This hub owns the licensing conversation across every modality. Where a specialist page exists, the link below goes there; where it does not, your request goes straight to intake. - the existing catalogue and its coverage - [Speech and audio inventory](https://ypai.ai/audio/datasets/) - recruitment by dialect, setting and device - [Speech collection programmes](https://ypai.ai/speech-data/) - collection at scale and the video families - [Video](https://ypai.ai/video-data/) - [Physical AI and sensor streams](https://ypai.ai/physical-ai-data/) - How your own data is processed when you bring it Each rail below is the medium. Select one and the record beside it fills in: what one record is for that medium, whose rights it carries, and which page has the detail. Where a specialist page exists, the link goes there; where it does not, your request goes to intake. dataset card, datasheet, machine-readable provenance and permitted use as the schedule sets them, the same for every medium - The existing catalogue and its coverage, by language, dialect, setting and device. - Recruitment by dialect, setting and device, when the catalogue has the gap. - Robot episodes, demonstrations and synchronised sensors, licensed or collected. - Collection at scale and the video families. - Scoped at intake against the requirement. - Scoped at intake, with source licence and redaction state. Multimodal bundles are scoped at intake, each stream with its own record. - one recording, its transcript and its speaker metadata - consent per speaker, per session · withdrawable - one sequence with its frame timing - [Video collection](https://ypai.ai/video-data/) - one frame or one synchronised capture with its calibration - recorded for the site and every person in frame - one document with its source, language and redaction state - the source licence and the consent basis, recorded - one episode with its demonstrations and synchronised streams - [Physical AI data](https://ypai.ai/physical-ai-data/) - one aligned bundle, each stream with its own record the record travels with the delivery ## Describe what you need. Start under NDA. Two sentences are enough: what data, for what use. A named person reads it and comes back with what exists, what can be sourced, and what would be collected. A named person reads it and comes back with what exists, what can be sourced, and what would be collected. ### Name - Yes, if the rights record holds I have read the Privacy Policy and consent to YPAI processing my data to answer this enquiry. ## Questions before the NDA. ### What does YPAI license, and what does it source? YPAI licenses rights-cleared datasets it owns, and it represents datasets from partner suppliers as curated dataset sourcing, with rights and availability checked before you see a sample. Where the data you need exists somewhere else, YPAI sources it to order. Where it does not exist yet, YPAI collects only that gap. The four forms sit on one line, from data YPAI already holds to data collected for you, and the same consent and licence trace applies to all of them. ### What can we see under NDA? Under NDA you see dataset cards, the inventory that matches your requirement, prices, provenance descriptions, technical specifications, anonymised or controlled samples, and partner and sourcing information where the partner agreement allows it. The NDA is the first gate, and it opens the inventory. Training rights, evaluation on production data, resale, sublicensing, white-label rights, access to personal data and keeping samples are granted by the licence that follows. ### What comes after the NDA? A Dataset Licence Agreement, or a Master Services Agreement with a Statement of Work and a Data Licence Schedule. The schedule names the dataset, the licensor authority, the permitted uses, the derivative rules, sublicensing, territory, term and exclusivity, and it records the consent version that was active when the data was made. Samples and evaluation material provided before signature are deleted at the request of YPAI as licensor once evaluation ends. ### How is a licensed record traceable to consent? Each record carries the agreement that authorised it, the consent version that was active, the rights that applied when it was created and the uses allowed now, together with its lineage and an integrity check. The same record answers a contributor's withdrawal: the affected records, datasets and deliveries are identified and the remediation procedure applies to your copy. The record ships with the delivery, in machine-readable dataset metadata, so your compliance team can read it directly. ### Can we have exclusivity, a limited territory, or a fixed term? Yes. Licences are structured exclusive or non-exclusive, term-based or renewing, with territory, permitted use and downstream-rights structures set in the schedule. Exclusivity is most often agreed for data collected to order, where the dataset is built against your requirement. The structure is agreed at scoping, together with the commercial terms. ### What if the data we need does not exist yet? Then that part is collected. YPAI names the gap against your requirement, missing languages, accents, age groups, scene or session types, formats, annotation or rights scope, and collects only that, under the same consent and rights controls as the inventory. Licensing what exists and collecting only the gap shortens the project and lowers the cost compared with collecting everything from zero. ### Can YPAI supply data through an intermediary? Yes, as white-label supply, where the end client does not necessarily contract with YPAI directly. That structure is written precisely: sublicensing, white-label rights, end-client disclosure, permitted use, model training, evaluation, derivative models, term, territory, exclusivity and indemnity are each set in the agreement. ### How do we start? Describe what you need in the form above: what data, for what use, which modality, whether existing inventory is acceptable, and whether an NDA is in place. A named person reads it and comes back with what exists, what can be sourced through partners, and what would be collected, and the NDA is put in place before any inventory or sample is shared. Rights-cleared data, with the licence attached. Describe what you need. YPAI comes back with what exists, what can be sourced, and what would be collected.