Genkit Documentation: Google's Open-Source Framework for Full-Stack AI and Agent Apps
by Google (Firebase)
Build typed AI flows, tool-calling agents, RAG and MCP integrations in TypeScript, Go, Dart or Python, with one API across model providers.
Overview
Genkit is an Apache 2.0 framework from Google's Firebase team for building full-stack AI applications, and its documentation at genkit.dev is the canonical guide. The GitHub README marks JavaScript/TypeScript and Go as production-ready, Python as Beta and Dart as Preview. The docs are organized into a get-started track (overview, setup, developer tools), a core-concepts track that covers flows, content generation, tool calling, Dotprompt prompt templating, runtime context, middleware, agentic patterns, interrupts, persistent chat, multi-agent systems, RAG, MCP, durable streaming, frontend integration, testing, evaluation and local observability, and a newer full-stack agents section. That section walks through defining agents, running and streaming turns, serving agents over HTTP, sessions and session stores, human-approval interrupts, background execution, multi-agent delegation, custom orchestration and generative UI through A2UI; the TypeScript Agents API is labeled Beta and Go's sits under an experimental package. Further sections cover backend frameworks (Express, FastAPI, Flask, Gin, NestJS and others), app frameworks (Next.js, Angular, Flutter, SvelteKit), model plugins for Google, OpenAI, Anthropic, Bedrock, Azure AI Foundry, xAI, DeepSeek, OpenRouter and Ollama, vector stores such as pgvector, Pinecone, Chroma and LanceDB, deployment to Firebase, Cloud Run, AWS Lambda or any container host, and OpenTelemetry-based monitoring that follows the GenAI semantic conventions. The repository has about 6.5k GitHub stars and shipped CLI/Dev UI 1.44.0 and Python SDK 0.12.0 releases in September and October.
At a Glance
- Topic
- Frameworks
- Level
- Intermediate
- Format
- Documentation
- Cost
- Free
- Duration
- Self-paced reference; ~1 hour for the get-started path, a few days to work through core concepts and the agents section
- Provider
- Google (Firebase)
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Define typed Genkit flows with structured output schemas and call them from your backend
- ✓Wire tool calling and interrupts so an agent pauses for human approval before acting
- ✓Build retrieval-augmented generation with Genkit retrievers over pgvector, Pinecone, Chroma or LanceDB
- ✓Connect agents to external tools and data through Genkit's Model Context Protocol support
- ✓Manage agent sessions, state snapshots and session stores for persistent multi-turn chat
- ✓Delegate work across specialized agents and write custom orchestration loops when the default loop falls short
- ✓Test and evaluate flows locally with the Genkit CLI and Developer UI traces
- ✓Deploy Genkit apps to Firebase, Cloud Run or AWS Lambda and export OpenTelemetry metrics
Highlights
- •One API across many model providers, so swapping Gemini for Claude or a local Ollama model is a plugin change
- •Covers four languages, with TypeScript and Go marked production-ready and Python and Dart in Beta or Preview
- •The full-stack agents section documents sessions, background execution and A2UI generative UI, which most framework docs leave out
- •Integration pages for a dozen backend frameworks and nine frontend frameworks, including FastAPI, Express, Next.js and Flutter
- •Built-in Developer UI for local tracing plus OpenTelemetry GenAI semantic conventions for production monitoring
- •Active project: about 6.5k GitHub stars and releases shipped in September and October
Who It’s For
Best For
- ✓TypeScript or Go developers adding LLM features and agents to an existing web or mobile backend
- ✓Firebase and Google Cloud teams that want a supported path to Cloud Run or Firebase deployment
- ✓Engineers who want provider-neutral model access without adopting a Python-first agent framework
- ✓Full-stack developers building chat or agent UIs on Next.js, Angular or Flutter
Prerequisites
- •Working knowledge of TypeScript/Node.js, Go, Dart or Python
- •An API key for at least one model provider (Gemini, OpenAI, Anthropic) or a local Ollama install
- •Basic familiarity with LLM concepts such as prompts, tool calling and embeddings
FAQ
What is Genkit Documentation: Google's Open-Source Framework for Full-Stack AI and Agent Apps?
The official documentation for Genkit, the open-source framework Google's Firebase team builds and runs in production for AI-powered and agentic applications. It is written for application developers who want one API over Gemini, OpenAI, Anthropic, Ollama and other models, and who need to ship flows, tool-calling agents and RAG into a real backend or web app.
Is Genkit Documentation: Google's Open-Source Framework for Full-Stack AI and Agent Apps free?
Genkit Documentation: Google's Open-Source Framework for Full-Stack AI and Agent Apps is free to access.
What level is Genkit Documentation: Google's Open-Source Framework for Full-Stack AI and Agent Apps for?
Genkit Documentation: Google's Open-Source Framework for Full-Stack AI and Agent Apps is aimed at a intermediate audience. Recommended background: Working knowledge of TypeScript/Node.js, Go, Dart or Python, An API key for at least one model provider (Gemini, OpenAI, Anthropic) or a local Ollama install, Basic familiarity with LLM concepts such as prompts, tool calling and embeddings.
How long does Genkit Documentation: Google's Open-Source Framework for Full-Stack AI and Agent Apps take?
Expect roughly Self-paced reference; ~1 hour for the get-started path, a few days to work through core concepts and the agents section. Most learners work through it at their own pace.
What will I learn from Genkit Documentation: Google's Open-Source Framework for Full-Stack AI and Agent Apps?
You'll learn: Define typed Genkit flows with structured output schemas and call them from your backend; Wire tool calling and interrupts so an agent pauses for human approval before acting; Build retrieval-augmented generation with Genkit retrievers over pgvector, Pinecone, Chroma or LanceDB; Connect agents to external tools and data through Genkit's Model Context Protocol support; Manage agent sessions, state snapshots and session stores for persistent multi-turn chat; Delegate work across specialized agents and write custom orchestration loops when the default loop falls short; Test and evaluate flows locally with the Genkit CLI and Developer UI traces; Deploy Genkit apps to Firebase, Cloud Run or AWS Lambda and export OpenTelemetry metrics.
Topics
Sources
This page was written from 4 sources, 2 on domains other than genkit.dev.