Covaric

Covaric is a research-driven company building controlled execution testing for AI agents. It records a baseline run of an agent, then re-runs it under simulated tool responses (errors, empty, partial, malformed, delayed) and forced decisions, without calling the tools again. A deterministic detector gives consistent verdicts, while a coverage report shows which steps, tools, response classes, and decisions were tested. Designed to run as a step in CI pipelines, it currently supports LangChain agents. The company reports strong precision and recall identifying behavioral failures across a benchmark of open-source agents, and is currently recruiting design partners.