Firestore vector store docs (#18)

* Firestore vector store docs

* indexing

* prettier

* feedback

docs/_guides.yaml

@@ -67,8 +67,6 @@ toc:
     path: /docs/genkit/plugins/pinecone
   - title: pgvector (code template)
     path: /docs/genkit/templates/pgvector
-  - title: Firestore vector store (code template)
-    path: /docs/genkit/templates/firestore-vector
   - title: Firebase
     path: /docs/genkit/plugins/firebase
   - title: Google Cloud

docs/index.md

@@ -370,6 +370,7 @@ maintained by the Genkit team:
     <td><code>firebase</code></td>
     <td>
       <b>Cloud deployment</b>: Cloud Functions, Firebase Authentication, App Check<br>
+      <b>Vector database</b>: Cloud Firestore vector store<br>
     </td>
   </tr>
 </table>

docs/plugins/firebase.md

@@ -2,6 +2,7 @@
 
 The Firebase plugin provides several integrations with Firebase services:
 
+- Indexers and retrievers using Cloud Firestore vector store
 - Trace storage using Cloud Firestore
 - Flow deployment using Cloud Functions
 - Authorization policies for Firebase Authentication users
@@ -58,7 +59,82 @@ Application Default Credentials. To specify your credentials:
 This plugin provides several integrations with Firebase services, which you can
 use together or individually.
 
-### Cloud Firestore
+### Cloud Firestore vector store
+
+You can use Cloud Firestore as a vector store for RAG indexing and retrieval.
+
+The `firebase` plugin provides a convenience function for defining Firestore
+retrievers, `defineFirestoreRetriever()`:
+
+```js
+import { defineFirestoreRetriever } from '@genkit-ai/firebase';
+import { initializeApp } from 'firebase-admin/app';
+import { getFirestore } from 'firebase-admin/firestore';
+
+const app = initializeApp();
+const firestore = getFirestore(app);
+
+const yourRetrieverRef = defineFirestoreRetriever({
+  name: 'yourRetriever',
+  firestore: getFirestore(app),
+  collection: 'yourCollection',
+  contentField: 'yourDataChunks',
+  vectorField: 'embedding',
+  embedder: textEmbeddingGecko,
+  distanceMeasure: 'COSINE', // 'EUCLIDEAN', 'DOT_PRODUCT', or 'COSINE' (default)
+});
+```
+
+To use it, pass it to the `retrieve()` function:
+
+```js
+const docs = await retrieve({
+  retriever: yourRetrieverRef,
+  query: 'look for something',
+  config: { limit: 5 },
+});
+```
+
+For indexing, use an embedding generator along with the Admin SDK:
+
+```js
+import { initializeApp } from 'firebase-admin';
+import { getFirestore, FieldValue } from 'firebase-admin/firestore';
+import { textEmbeddingGecko } from '@genkit-ai/vertexai';
+import { embed } from '@genkit-ai/ai/embedder';
+
+const app = initializeApp();
+const firestore = getFirestore(app);
+
+const indexConfig = {
+  collection: 'yourCollection',
+  contentField: 'yourDataChunks',
+  vectorField: 'embedding',
+  embedder: textEmbeddingGecko,
+};
+
+async function indexToFirestore(content) {
+  const embedding = await embed({
+    embedder: indexConfig.embedder,
+    content,
+  });
+  await firestore.collection(indexConfig.collection).add({
+    [indexConfig.vectorField]: FieldValue.vector(embedding),
+    [indexConfig.contentField]: content,
+  });
+}
+```
+
+Firestore depends on indexes to provide fast and efficient querying on
+collections. The prior example requires the `embedding` field to be indexed to
+work. To do so, invoke the function and Firestore will throw an error with a
+command to create an index. Execute that command and your index should be ready
+to use.
+
+See the [Retrieval-augmented generation](../rag.md) page for a general
+discussion on indexers and retrievers.
+
+### Cloud Firestore trace storage
 
 You can use Cloud Firestore to store traces:
 

docs/rag.md

@@ -249,6 +249,7 @@ export const ragFlow = defineFlow(
 Genkit provides indexer and retriever support through its plugin system. The
 following plugins are officially supported:
 
+- [Cloud Firestore vector store](plugins/firebase.md)
 - [Chroma DB](plugins/chroma.md) vector database
 - [Pinecone](plugins/pinecone.md) cloud vector database
 
@@ -257,7 +258,6 @@ code templates, which you can customize for your database configuration and
 schema:
 
 - PostgreSQL with [`pgvector`](templates/pgvector.md)
-- [Firestore vector store](templates/firestore-vector.md)
 
 Embedding model support is provided through the following plugins: