Ozperium has released AgentSpec, a testing framework designed for non-deterministic AI agents. The tool allows developers to write YAML-based tests with assertions tailored to AI output, including semantic similarity, JSON path validation, latency checks, and LLM-as-judge evaluation using local models. AgentSpec supports CI integration through JUnit XML output, GitHub Actions, and behavior regression reports that highlight output changes between runs. The framework is MIT-licensed and available via npm.
No score is assigned. Sources and their independence are shown in the citation chain below.