Plugins overview

Extend Genkit with official and community plugins for model providers, vector stores, telemetry, and more.

Genkit’s plugin system lets you swap in any model provider, vector store, or telemetry backend without changing your application code. A plugin registers one or more actions — models, embedders, retrievers, indexers, evaluators — into the Genkit registry. Your flows call those actions by name and never depend on a specific provider’s SDK directly.

How plugins extend Genkit

Every plugin follows the same lifecycle:

1. Register — pass the plugin to genkit({ plugins: [...] }) (JS/TS), genkit.Init(ctx, genkit.WithPlugins(...)) (Go), or Genkit(plugins=[...]) (Python).
2. Lazy init — the registry calls init() / Init() on first use, not at startup, so cold-start cost is minimised.
3. Resolve — when your code references googleai/gemini-2.5-flash, the registry asks the googleai plugin to materialise that action on demand.
4. List — the Dev UI calls list() / list_actions() to enumerate every available model or embedder for the action explorer.

You can configure multiple plugins in the same Genkit instance — for example googleAI() for prototyping and vertexAI() for production — and choose between them at call time by qualifying the model name with its provider prefix.

Official plugins

JavaScript / TypeScript

Plugin	Package	Category	Notes
Google AI (Gemini)	@genkit-ai/google-genai	Model provider	Gemini, Imagen, Veo, TTS via AI Studio API
Vertex AI	@genkit-ai/google-genai	Model provider	Gemini + Imagen on GCP (same package, different export)
Ollama	@genkit-ai/ollama	Model provider	Local models via Ollama server
Firebase	@genkit-ai/firebase	Vector store / telemetry	Firestore vector search, Firebase telemetry
Anthropic	genkitx-anthropic	Model provider	Claude 3.5 / 4 (community)
OpenAI-compatible	genkitx-openai	Model provider	OpenAI, OpenRouter, together.ai (community)
Pinecone	genkitx-pinecone	Vector store	Pinecone vector database (community)
Chroma	genkitx-chromadb	Vector store	ChromaDB (community)
LangChain	genkitx-langchain	Integration	Use LangChain tools inside Genkit (community)
Google Cloud	@genkit-ai/google-cloud	Telemetry	Cloud Trace + Cloud Logging

Go

Plugin	Import path	Category
Google AI / Vertex AI	github.com/firebase/genkit/go/plugins/googlegenai	Model provider
Ollama	github.com/firebase/genkit/go/plugins/ollama	Model provider
Firebase	github.com/firebase/genkit/go/plugins/firebase	Vector store / telemetry
Google Cloud	github.com/firebase/genkit/go/plugins/googlecloud	Telemetry

Python

Plugin	Package	Category
Google AI (Gemini)	genkit-google-genai-plugin	Model provider
Vertex AI	genkit-google-genai-plugin	Model provider
Ollama	genkit-ollama-plugin	Model provider
Anthropic	genkit-anthropic-plugin	Model provider
OpenAI-compatible	genkit-compat-oai-plugin	Model provider
DeepSeek	genkit-deepseek-plugin	Model provider
xAI (Grok)	genkit-xai-plugin	Model provider
Mistral	genkit-mistral-plugin	Model provider
Hugging Face	genkit-huggingface-plugin	Model provider
Firebase	genkit-firebase-plugin	Vector store / telemetry
Google Cloud	genkit-google-cloud-plugin	Telemetry

Quick start

TypeScript

import { genkit } from 'genkit';
import { googleAI } from '@genkit-ai/google-genai';

const ai = genkit({
  plugins: [googleAI()],
});

const { text } = await ai.generate({
  model: googleAI.model('gemini-2.5-flash'),
  prompt: 'Hello, Genkit!',
});

Go

import (
  "github.com/firebase/genkit/go/genkit"
  "github.com/firebase/genkit/go/plugins/googlegenai"
)

g := genkit.Init(ctx,
  genkit.WithPlugins(&googlegenai.GoogleAI{}),
)

resp, _ := genkit.Generate(ctx, g,
  ai.WithModelName("googleai/gemini-2.5-flash"),
  ai.WithPrompt("Hello, Genkit!"),
)

Python

from genkit import Genkit
from genkit.plugins.google_genai import GoogleAI

ai = Genkit(plugins=[GoogleAI()])

response = await ai.generate(
    model='googleai/gemini-2.5-flash',
    prompt='Hello, Genkit!',
)

The plugin interface

At its core, a plugin implements three methods:

Method	Purpose
init()	One-time setup — create API clients, register known actions. Called lazily on first use.
resolve(actionType, name)	Materialise a single action by kind and name (e.g. a model not registered at init time).
list()	Return metadata for all available actions so the Dev UI can enumerate them.

In TypeScript the modern interface is GenkitPluginV2 (from genkit/plugin). In Go plugins implement the api.Plugin interface. In Python plugins extend the abstract Plugin base class from genkit._core.plugin.

See Writing plugins for a complete walkthrough.

Related pages

Google AI plugin

Gemini, Imagen, Veo, and embeddings via the Gemini Developer API.

Vertex AI plugin

Enterprise access to Google models on Google Cloud.

Ollama plugin

Run Llama, Mistral, Gemma, and other models fully locally.

Firebase plugin

Firestore vector search and Firebase telemetry.

Writing plugins

Build your own plugin and publish it to npm.

RAG guide

Use retrievers and embedders together for retrieval-augmented generation.