“research”

Simulated customer-service benchmark where an agent follows domain policies and uses tools while a simulated user also acts.

Benchmark and framework that tests models and agents on tasks using real-world MCP servers.