Summary

firebase/genkit is an OSS framework for building AI-powered apps in JS, Go, and Python. Used in production by Google’s BaaS, Firebase.

Create a Genkit Go plugin that integrates MongoDB Atlas Vector Search as a vector store for RAG applications, implementing the ai.Embedder and ai.Retriever interfaces. This would enable Genkit users to store embeddings and perform similarity searches using MongoDB, filling a gap in the current plugin ecosystem which includes Pinecone, Weaviate, and PostgreSQL but lacks MongoDB support. The integration increases Go driver visibility in the growing AI/LLM application space while providing Atlas users a seamless path to building RAG features.

Definition of Done

• Create a mongovector plugin that Implement ai.Embedder and ai.Retriever interfaces
• Add a mongovector sample package with usage
• Track usage of the plugin through client metadata

The example API would likely look something like this, following existing plugin patterns:

// Compile-time interface checks.
var _ ai.Retriever = (*Retriever)(nil)

// DocStore wraps a MongoDB collection for vector search operations.
type DocStore struct {}

// Retriever implements ai.Retriever for MongoDB Atlas Vector Search.
type Retriever struct {
ds   *DocStore
}

// DefineRetriever creates and registers a Genkit retriever for similarity search.
func DefineRetriever(ds *DocStore, name string, opts RetrieverOptions) *Retriever

// Name returns the retriever name.
func (r *Retriever) Name() string

// Retrieve performs similarity search against MongoDB Atlas Vector Search.
func (r *Retriever) Retrieve(ctx context.Context, req *ai.RetrieverRequest) (*ai.RetrieverResponse, error)

// Register registers the retriever with the Genkit registry.
func (r *Retriever) Register(reg ai.Registry)