Genkit

Firebase's team at Google built Genkit as an open source framework for AI applications, and Google runs it in production, which distinguishes it from most vendor SDKs. Code is organized around flows, observable typed functions that wrap model calls, tool use, and retrieval so they can be traced, tested, and deployed as units, and a local developer UI lets you run flows, inspect traces, and evaluate outputs before shipping. TypeScript and Go SDKs are stable while Python and Dart are in preview, and a plugin system connects Gemini, Vertex AI, OpenAI, Anthropic, and local models through Ollama, alongside vector stores for retrieval augmented generation. Deployment targets include Cloud Functions, Cloud Run, or any Node or Go host, with framework integrations for Next.js, Express, and others. The Apache-2.0 repository sits around 6,300 GitHub stars, and the framework itself is free with costs coming only from whichever model provider you attach. Teams already on Firebase or Google Cloud pick it for structured, observable AI features without committing to a single model vendor.