nnull13/gaugo

Gaugo is a Go library for evaluating AI applications such as RAG systems (retrieval-augmented generation, meaning AI that answers questions using a knowledge base), chatbots, and agents. Its main selling point is that it fits naturally into Go's standard testing workflow: you write evaluation cases as ordinary Go tests and run them with go test, which means AI evaluations can live alongside application code and run in any CI pipeline that already runs Go tests. The framework offers two kinds of checks. Deterministic checks run without an LLM and verify things like whether a response contains a required phrase, matches a regex pattern, stays within a length limit, or responds within a latency budget. LLM judged metrics use a language model as a grader to assess qualities like context relevancy, faithfulness, meaning whether the answer sticks to the provided documents, answer relevancy, citation accuracy, and safety. You provide an adapter function that connects Gaugo to your application, and the framework calls it for each test case, collects results, and fails the test suite if any metric falls below its configured threshold. Gaugo includes 24 built-in metrics spanning RAG, safety, generation quality, structured output, instruction following, domain-specific checks, and deterministic contracts. Provider adapters are available for OpenAI, Anthropic, Gemini, xAI, and local model services. Test cases run concurrently by default while preserving registration order in results. The README positions Gaugo as a Go-native alternative to Python-first evaluation tools such as Ragas, DeepEval, and TruLens, for teams who want evaluations to live in their Go codebase.