For AI teams
Training data for long-horizon task agents
Short answer
Long-horizon task agents need records of work spanning days or weeks. SourceX sources these from businesses in the U.S., Canada, the EU, the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia that own the records, with personal details removed and every release approved by the supplier.
Useful records#
- Projects from kickoff to delivery
- Multi-week support cases
- Deal cycles in CRM
Why real business records help#
- Shows long dependencies
- Captures state over time
- Includes outcomes at the end
How a request works#
- Describe the records, volume and uses you need
- SourceX matches suitable suppliers in the U.S., Canada, the EU, the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia and runs a rights review
- A prepared sample is shared under NDA after supplier approval
- The license sets allowed uses, term and deletion; resale and re-identification are barred
Frequently asked questions
What records of work spanning days or weeks are used to train long-horizon task agents?
Long-horizon task agents are trained on records of work spanning days or weeks. Useful records include projects from kickoff to delivery, multi-week support cases, and deal cycles in CRM. They show long dependencies, state captured over time, and the outcomes at the end of each cycle.
What gets removed from projects, support cases and CRM deal histories before release?
Records for long-horizon task agents have personal details removed, and the supplier approves each release. A rights review runs on projects, support cases and CRM deal histories before a sample is prepared. The sample is shared under NDA only after supplier approval.
How does a request for long-horizon task agent data work, from scope to license?
Long-horizon task agents are scoped through a request describing the records, volume and uses you need. SourceX matches suitable suppliers in the U.S., Canada, the EU, the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia and runs a rights review. A prepared sample is shared under NDA after supplier approval, and the license sets allowed uses, term and deletion.
More in Datasets
- Computer-use agents
- Document understanding models
- Agentic workflow planning
- Domain-specific fine-tuning
- IT operations agents
- Email and calendar agents
- Multilingual enterprise models
- Multimodal document models
- Process supervision
- Recruiting AI
- Reward models
- RL environments
- Safety and red teaming
- Tool use and function calling