Google launches Firebase Genkit, a new open source framework for building AI-powered apps

At the Google I/O developer conference on Tuesday, Google announced an addition to its Firebase platform that aims to make it easier for developers to build AI-powered applications in JavaScript/TypeScript, with Go support coming soon.

Firebase Genkit is an open source framework, using the Apache 2.0 license, that enables developers to quickly build AI into new and existing applications.

Some of the use cases for Genkit that the company is highlighting Tuesday include many of the standard generative AI use cases: content generation and summarization, text translation and generating images.

“Powerful large language models put AI-powered app features within reach, but building and refining these features beyond a prototype is hard,” Google product manager Chris Gill and developer advocate Peter Friese wrote in Tuesday’s announcement. “Many of us are still figuring out how to deploy these features in production at scale and understand how they are performing so we can quickly iterate and improve on them. Add to this the need to balance safety and stability throughout the process and the problem becomes even harder. Let’s face it, everyone can use some help.”

The Firebase team promises that developers will be able to jump right into using Genkit because it uses the same approaches as the rest of the Firebase toolchain. Using Genkit, they’ll be able to test their new features locally and then deploy their application with the help of Google’s serverless platforms like Cloud Functions for Firebase and Google Cloud Run.

Because it’s open source, developers will be able to extend Genkit as needed, but out of the box, it already supports a number of third-party open source projects. That means that on top of Google’s own Gemini models, for example, developers can use open models via Ollama. Genkit will also support vector databases like Chrome, Pinecone and PostgreSQL’s pgvector, in addition to Google Cloud Firestore.

“Genkit is also designed to be open to any and all models, vector stores, embedders, evaluators, and other components through its plugin system,” the team writes.