01Training / Evaluation
Agentic coding
Repository-backed tasks for software-engineering agents, with executable tests, environment state, and patch-level evaluation.
Training data & evaluation environments
Data for intelligence that works.
Realistic tasks. Executable environments. Verifiable outcomes.
Training and evaluation data for agents and multimodal systems.
Define the frontier.
Move intelligence forward.
San Francisco / First lightMorrowLab
A new day for intelligence
37°49′ NBuild toward the next horizon.
122°28′ W
Designed around real work
Agentic coding / Computer use / Multimodal systems
Scroll to explore ↓Data categories / Capabilities
Data for real-world capabilities.
Data spanning agentic coding, automated research, computer use, multimodal learning, and embodied AI.
02Training / Evaluation
Automated research
Kernel-optimization environments with correctness gates and latency and memory objectives for systems research.
03Training / Evaluation
Computer-use agents
Stateful cross-application workflows evaluated through artifact validity, evidence traceability, and rubric-based completion.
04Training / Evaluation
Long-horizon environments
Persistent workspaces with multi-stage dependencies, tool use, failure recovery, and end-state verification.
05Training / Evaluation
Multimodal corpora
Rights-scoped audiovisual and image-transformation data for representation learning, perception, and generation.
06Training / Evaluation
Embodied AI & world models
Synchronized egocentric video, pose, and motion data for action understanding, visuomotor learning, and world-model research.
Explore sample catalogs · Request access
Need a closer look?