Environment
Define the work
The environment owns the agent's tools, actions, constraints, and feedback. It makes the task concrete—and defines what useful intelligence must accomplish.
- Tasks + action space
- Tools + constraints
- Feedback + training data
Benchmark
Quantify intelligence
Measure performance inside the environment, not in the abstract. Own the benchmarks that reflect your work, publish them when useful, and earn when others run them.
- Task-specific evaluation
- Model comparisons
- Publish + earn
Router
Choose the intelligence
Use one interface for open and closed models. Route each task to the model proven to perform it best, balancing quality, latency, privacy, and cost.
- Open + closed models
- Evidence-based routing
- Costs + traces