Licensed data.Prepared for AI.
We connect the organisations that hold valuable data with the AI developers who need it. From publishing archives and footage to code and workplace records, we source, prepare and license datasets for defined AI applications.
Rights-aware sourcing
Authority and permitted use are assessed before data is prepared.
Purpose-specific licensing
Each licence is tied to defined uses and restrictions.
Structured preparation
Data is scoped, filtered and organised to fit a stated purpose.
Documented provenance
Each dataset is accompanied by a record of where it came from and how it was handled.
Valuable data already exists. Access is the challenge.
Publishers, archives, production companies and businesses hold large bodies of material AI developers need: specialist text, original footage, code and the record of how work actually gets done.
AI developers need exactly this kind of realistic, contextual data, with clear provenance and agreed terms of use. We act as the commercial and operational link between the two, so each side doesn't have to become an expert in the other.
Existing data
Text, footage, code and records held by publishers, archives and businesses. Often rich in context, rarely organised for outside use.
Assessed and prepared
Rights and permitted uses are reviewed by people, not assumed. Suitable data is scoped, filtered, de-identified where appropriate and documented. Unsuitable data is left out.
Human review in every engagement
Licensed for approved uses
AI developers receive defined datasets under licences that state what the data may be used for, and what it may not.
For data partners
Your data may be worth more than you think.
Archives of text, footage, code and records can be valuable to AI developers. If your organisation holds suitable material and has the authority to license it, we assess it, find the right buyer and structure the licence. You don't need any AI capability of your own.
- Explore the commercial value of existing data assets.
- Understand which datasets may fit AI development demand.
- Establish clear licensing terms and permitted uses.
- Explore potential ongoing data partnerships.
What starting involves
- A short enquiry and a conversation. No data changes hands.
- A preliminary assessment based on descriptions, not raw records.
- A decision on whether to proceed, made by you.
For AI companies
Find the data your models need.
Tell us what your model or agent needs to learn from, or be evaluated against. We source and prepare licensed datasets that match, with documented provenance and licence terms built around your intended use.
- Specify the data, structure and scale you need.
- Get a clear view of feasibility before either side invests.
- Discuss data quality and provenance requirements.
- Establish licensing arrangements around intended use.
- application
- Coding-agent evaluation
- source
- Software company archive, licensed
- context
- Issue, review thread, merged change
- volume
- ~20,000 resolved issues
- languages
- Python, TypeScript
- format
- JSONL, one issue thread per line
- identifiers
- Names, emails and secrets removed
- licence
- Evaluation only, no redistribution
Data categories
Data from the organisations that create it.
The data we source for AI teams, licensed from organisations with the authority to license it. Select a category to see typical sources, the context that can be preserved and what needs care.
Sourced to brief
Publishing and text archives
- Typical sources
- Specialist and academic publishers, learned societies, trade-journal owners and archive holders
- Context that could be preserved
- Full text with its structure: chapters, sections, references
- Bibliographic and subject metadata
- Editorial and revision history
- A confirmed rights chain for each title
- Potential uses
- Pre-training, fine-tuning, domain grounding and evaluation.
- Considerations
- Digital and AI rights often sit with authors under older contracts. The licence chain for each title is confirmed before anything is offered.
Sourced to brief
Video and footage archives
- Typical sources
- Production companies, broadcast and educational archives, drone and specialist camera operators
- Context that could be preserved
- Original resolution and frame rate
- Shot and scene metadata
- Linked audio tracks and transcripts
- Capture details such as setting and equipment
- Potential uses
- Video generation, world models and video understanding.
- Considerations
- Owning the footage is not enough on its own. Identifiable people, private locations and music in shot each need clearing.
Sourced to brief
Software and engineering records
- Typical sources
- Software companies, including businesses that are winding down, restructuring or have been acquired
- Context that could be preserved
- Links between issue, discussion and change
- Review comments in sequence
- Test suites and build history
- Incident timelines and outcomes
- Potential uses
- Coding agents, code review assistance and agent environments with checkable outcomes.
- Considerations
- Code is checked for third-party and open-source licence obligations, secrets and credentials, and any customer data before it can be licensed.
Sourced to brief
Workplace records
- Typical sources
- Collaboration, email, ticketing and project systems, typically from organisations that are closing or restructuring
- Context that could be preserved
- Thread structure and reply order
- Handoffs and approvals between roles
- Linked tickets, documents and decisions
- Roles rather than identities
- Potential uses
- Realistic agent environments, multi-turn dialogue and workflow evaluation.
- Considerations
- Employees and outside correspondents are identifiable people, so the notice they were given, the lawful basis and confidentiality owed to clients decide what is possible. Platform terms can also restrict how data is extracted.
Sourced to brief
Customer support records
- Typical sources
- Ticketing systems, live chat and help-desk platforms
- Context that could be preserved
- Issue category and priority
- Resolution path and escalations
- Agent and customer turns
- Outcome and satisfaction signals
- Potential uses
- Support agents, intent classification and resolution-quality evaluation.
- Considerations
- Contains customer personal data. The terms and privacy notices given to customers shape what is possible.
Sourced to brief
Documentation and knowledge bases
- Typical sources
- Wikis, intranets, product documentation and document repositories
- Context that could be preserved
- Document structure and hierarchy
- Revision history
- Cross-references
- Questions the content answers
- Potential uses
- Retrieval, enterprise question answering and grounding evaluation.
- Considerations
- Frequently confidential. Ownership of contributed and third-party content is confirmed first.
Sourced to brief
Expert and professional input
- Typical sources
- UK-qualified professionals in law, finance, accounting and regulated sectors, engaged against a defined brief
- Context that could be preserved
- Credentials and specialism of each contributor
- Reasoning alongside final answers
- Rubrics and reviewer agreement
- Rights assigned by contract
- Potential uses
- Fine-tuning, evaluation sets, preference data and domain benchmarks.
- Considerations
- Quality control and conflict checks matter more than volume. Contributors must not draw on client-confidential information.
Sourced to brief
Speech and conversational audio
- Typical sources
- Recording projects with consenting speakers, arranged to a buyer's specification
- Context that could be preserved
- Speaker metadata such as accent and region
- Verified transcripts and timestamps
- Speaker turns and overlaps
- Recording conditions
- Potential uses
- Speech recognition, voice agents and speech synthesis.
- Considerations
- Every speaker gives informed consent for the stated uses. Voice is personal data, and biometric data when used to identify someone.
How it works
Each engagement follows the same five stages. Several of them depend on people reviewing rights, context and suitability, so timelines vary with the data and the parties involved.
Define the opportunity
Understand the data that is available, or the AI company's requirements, in enough detail to judge fit.
Establish the rights and scope
Assess ownership, authorisation, privacy, confidentiality and the purposes for which the data may be used.
Human review
Prepare the dataset
Where appropriate and authorised, extract, structure, filter, document and quality-check the relevant data.
Agree the licence
Define the specific dataset, permitted uses, restrictions, delivery arrangements and commercial terms.
Deliver and develop
Provide the agreed dataset securely and explore further supply or licensing where it makes sense for both sides.
Trust starts with the data.
Privacy, permissions and provenance are not add-ons to this model. They decide whether a dataset can exist at all.
The specific controls for any dataset depend on the data itself, the jurisdictions involved and the intended use, and are agreed for each engagement.
Authorised data access
We work only with data that an organisation has the authority to share, through access methods agreed in writing.
Assessment of licensing rights
Contracts, notices and the original purposes of collection are reviewed to establish what uses could be permitted.
Data minimisation
Datasets are scoped to what a defined use requires. Fields and records outside that scope are excluded.
De-identification where appropriate
Direct identifiers are removed or replaced, and residual re-identification risk is assessed for the specific dataset.
Confidentiality protections
Commercially sensitive material is identified and excluded or restricted, and confidentiality terms apply on both sides.
Provenance and permitted-use records
Each dataset carries documentation of its source, processing and the uses its licence allows.
Security and access controls
Access to data in preparation is limited to the people who need it, and delivery uses agreed secure methods.
What we don't claim
- Removing names is not the same as anonymisation. Re-identification risk is assessed, not assumed away.
- De-identified data can still be personal data, and legal obligations may continue to apply.
- Holding data does not automatically mean it can be licensed. Some datasets won't be suitable.
Let's put the right data to work.
Whether you're exploring a data partnership or looking for datasets for AI development, start with a conversation.