Slack · Microsoft Teams · Google Chat
For AI labs
Real business records, licensed at the source.
We license de-identified records directly from the companies that created them: email threads, support tickets, sales histories, project work and code history. Rights are checked, structure stays intact and every dataset comes with clear terms.
What we supply
How work actually gets done.
Connected records from established operating companies, each showing a problem, the steps people took, the judgment calls along the way and the result.
Gmail · Outlook
Business email
Salesforce · HubSpot
CRM & sales records
Zendesk · Intercom
Customer support
Notion · Google Drive · SharePoint
Documents & guides
Zoom · Approved business calls
Call transcripts
Asana · Jira · ServiceNow
Projects & tasks
GitHub · GitLab
Code & engineering
QuickBooks · Xero
Finance & operations
Why it's different
Data your models haven't seen.
Licensed at the source
Each contributing company signs a licence covering scope, permitted uses and buyer restrictions, and confirms it holds the rights. You get documented provenance, not scraped text.
Structure intact
Threads, consistent pseudonyms for participants, timestamps, reply order and outcomes are preserved, so a model can learn cause and effect across a whole case.
Never on the public web
Private operating records that no crawler has reached. Useful for training, and for evaluation sets that can't have leaked into pretraining data.
Licensed uses
Training, tuning and evaluation.
- Training
- Real conversations and workflows that teach general skills: following up, resolving disputes, routing approvals.
- Fine-tuning
- Examples from specific kinds of work, such as support, B2B sales or finance operations.
- Evaluation
- Held-out cases your models have never seen, scored against real outcomes.
- Synthetic data
- Real records to seed and ground generated examples.
- Delivery
- De-identified by a specialist partner, then delivered to your controlled environment.
- Terms
- Training and evaluation only. No publication or resale of raw data, and contributing companies are not named.
Data requests
Tell us what your models need.
Share the domains, record types, volumes and timelines you're working to. We'll reply with what we can source and on what terms.