// the find
genkit-ai/genkit
Open-source framework for building AI-powered apps in JavaScript, Go, and Python, built and used in production by Google
Genkit is Google's framework for building LLM-powered applications in TypeScript, Go, Python, and Dart. It abstracts model providers behind a unified API and wraps the whole thing in flows — typed, observable units of AI logic. Aimed at backend developers who want structured output, RAG, and tool-calling without wiring everything from scratch.
The local developer UI with trace inspection is genuinely useful — you can see exactly what went into and out of each step of a flow, which is the thing that hurts most when debugging LLM pipelines. The plugin architecture is well thought out: swapping Gemini for Ollama or OpenAI is a config change, not a rewrite. Go support is first-class and production-ready, not an afterthought, which is rare in this space. DotPrompt (prompt templating with typed schemas) keeps prompt logic out of application code and makes it testable.
The Python SDK is still beta, so if that's your stack you're accepting instability on an otherwise mature framework. Evaluation tooling exists but is local-file-based — no built-in integration with external eval platforms, so serious eval pipelines need custom work. The framework leans heavily into the Firebase/Google Cloud deployment story; deploying to AWS or Azure without any Google services means giving up some monitoring and convenience integrations. Dart support is in preview with no clear timeline to stable, so mobile-first teams should treat it as experimental.