Backed by Y Combinator

Real-world data for agents that do real work

Each one is a digital twin of a real company: licensed, anonymized and consistent across every tool.

Datasets

Complete company histories, linked across every tool

Years of emails, chat, documents, tickets, code and CRM from real companies, anonymized with one entity dictionary so people, projects and clients stay consistent from one source to the next.

  • Multimodal: text, documents, spreadsheets, code, org charts
  • Cross-tool workflows preserved end to end
  • Licensed for training, code-named, PII removed
One workflow, six tools Retailer 3 · same entity in every tool
  1. Customer email “Order #4821 is short 4,000 units.” · Retailer 3 Gmail
  2. Support ticket TCK-219 opened for Retailer 3 Support
  3. Slack thread #ops · “Can we split Retailer 3’s shipment?” Slack
  4. Jira issue OPS-512 · Split shipment for Retailer 3 Jira
  5. Pull request PR #88 · Fix allocation for Retailer 3 GitHub
  6. CRM update Retailer 3 · account updated HubSpot

Digital Twins

Every dataset is a digital twin of a real company

An anonymized replica of a real company's tools and data. Names, dates and proprietary content change; structure and relationships stay, so your agents learn from how the company actually works.

  • Multi-tool coherence across every source
  • Structural fidelity: real volumes, real mess
Catalogue

Dataset samples

Every dataset is code-named. Request a sample to see the structure by reaching out to contact@ooakdata.com.

Trust

Safe by design

Licensed, not sold

Every dataset comes from a signed, training-only license with the company.

Anonymized in-house

Our own engine. No third party touches raw data.

Zero re-identification

Code-named datasets, with no incident to date.

Raw data deleted

Within 30 days of processing, as a contract obligation.

Tell us what your agents need to learn

We will scope the right dataset with you.

Get in touch