Testing & evals

Automated attacks on a chat agent to find security and quality failures.

The European red-team lab — attacking agents for a living, publishing the benchmarks and the incident datasets in the open, and selling the same machinery back to banks and insurers as a compliance instrument.

Simulated users that test an agent before launch, and monitoring after.

The test-harness layer of the agent stack — the people trying to make "does this agent actually work" a number you can put in CI, priced per event in euros and open enough to run on your own metal.