feat: Qdrant vector store

.env.local.example

@@ -1,6 +1,7 @@
 # Pick Vector DB
 VECTOR_DB=pinecone
 # VECTOR_DB=supabase
+# VECTOR_DB=qdrant
 
 # Clerk related environment variables
 NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=pk_****
@@ -34,4 +35,9 @@ TWILIO_ACCOUNT_SID=AC***
 TWILIO_AUTH_TOKEN=*****
 
 # Steamship related environment variables
-STEAMSHIP_API_KEY=****
+STEAMSHIP_API_KEY=****
+
+# Qdrant related environment variables
+QDRANT_URL="httpS://****"
+QDRANT_API_KEY=****
+QDRANT_COLLECTION_NAME=https://****

README.md

@@ -31,7 +31,7 @@ The stack is based on the [AI Getting Started Stack](https://github.com/a16z-inf
 
 - Auth: [Clerk](https://clerk.com/)
 - App logic: [Next.js](https://nextjs.org/)
-- VectorDB: [Pinecone](https://www.pinecone.io/) / [Supabase pgvector](https://supabase.com/docs/guides/database/extensions/pgvector)
+- VectorDB: [Pinecone](https://www.pinecone.io/) / [Supabase pgvector](https://supabase.com/docs/guides/database/extensions/pgvector) / [Qdrant](https://qdrant.tech/)
 - LLM orchestration: [Langchain.js](https://js.langchain.com/docs/)
 - Text model: [OpenAI](https://platform.openai.com/docs/models), [Replicate (Vicuna13b)](https://replicate.com/replicate/vicuna-13b)
 - Text streaming: [ai sdk](https://github.com/vercel-labs/ai)
@@ -93,8 +93,7 @@ c. **Replicate API key**
 
 Visit https://replicate.com/account/api-tokens to get your Replicate API key if you're using Vicuna for your language model.
 
-
-❗ **_NOTE:_** By default, this template uses Pinecone as vector store, but you can turn on Supabase pgvector easily by uncommenting `VECTOR_DB=supabase` in `.env.local`. This means you only need to fill out either Pinecone API key _or_ Supabase API key.
+❗ **_NOTE:_** By default, this template uses Pinecone as a vector store, but you can switch to Supabase pgvector or Qdrant by uncommenting `VECTOR_DB=supabase` or `VECTOR_DB=qdrant` in `.env.local`. This means you only need to fill out either `PINECONE_API_KEY`, `SUPABASE_API_KEY`, or Qdrant API details such as `QDRANT_URL`, `QDRANT_API_KEY`, and `QDRANT_COLLECTION_NAME`.
 
 d. **Pinecone API key**
 
@@ -148,6 +147,12 @@ npm run generate-embeddings-pinecone
 npm run generate-embeddings-supabase
 ```
 
+#### If using Qdrant
+
+```bash
+npm run generate-embeddings-qdrant
+```
+
 ### 5. Run app locally
 
 Now you are ready to test out the app locally! To do this, simply run `npm run dev` under the project root.

package.json

@@ -9,13 +9,15 @@
     "lint": "next lint",
     "generate-embeddings-pinecone": "node src/scripts/indexPinecone.mjs",
     "generate-embeddings-supabase": "node src/scripts/indexPGVector.mjs",
+    "generate-embeddings-qdrant": "node src/scripts/indexQdrant.mjs",
     "export-to-character": "node src/scripts/exportToCharacter.mjs"
   },
   "dependencies": {
     "@clerk/clerk-sdk-node": "^4.10.12",
     "@clerk/nextjs": "^4.21.9-snapshot.56dc3e3",
     "@headlessui/react": "^1.7.15",
     "@pinecone-database/pinecone": "^0.1.6",
+    "@qdrant/js-client-rest": "^1.9.0",
     "@supabase/supabase-js": "^2.25.0",
     "@tailwindcss/forms": "^0.5.3",
     "@types/node": "20.2.5",

src/app/utils/memory.ts

@@ -3,7 +3,9 @@ import { OpenAIEmbeddings } from "langchain/embeddings/openai";
 import { PineconeClient } from "@pinecone-database/pinecone";
 import { PineconeStore } from "langchain/vectorstores/pinecone";
 import { SupabaseVectorStore } from "langchain/vectorstores/supabase";
+import { QdrantVectorStore } from "langchain/vectorstores/qdrant";
 import { SupabaseClient, createClient } from "@supabase/supabase-js";
+import { QdrantClient } from '@qdrant/js-client-rest';
 
 export type CompanionKey = {
   companionName: string;
@@ -14,13 +16,16 @@ export type CompanionKey = {
 class MemoryManager {
   private static instance: MemoryManager;
   private history: Redis;
-  private vectorDBClient: PineconeClient | SupabaseClient;
+  private vectorDBClient: PineconeClient | SupabaseClient | QdrantClient;
 
   public constructor() {
     this.history = Redis.fromEnv();
     if (process.env.VECTOR_DB === "pinecone") {
       this.vectorDBClient = new PineconeClient();
-    } else {
+    } else if (process.env.VECTOR_DB === "qdrant") {
+      this.vectorDBClient = new QdrantClient({ url: process.env.QDRANT_URL!, apiKey: process.env?.QDRANT_API_KEY });
+    }
+    else {
       const auth = {
         detectSessionInUrl: false,
         persistSession: false,
@@ -58,6 +63,21 @@ class MemoryManager {
         { pineconeIndex }
       );
 
+      const similarDocs = await vectorStore
+        .similaritySearch(recentChatHistory, 3, { fileName: companionFileName })
+        .catch((err) => {
+          console.log("WARNING: failed to get vector search results.", err);
+        });
+      return similarDocs;
+    } else if (process.env.VECTOR_DB === "qdrant") {
+      console.log("INFO: using Qdrant for vector search.");
+      const qdrantClient = <QdrantClient>this.vectorDBClient;
+
+      const vectorStore = await QdrantVectorStore.fromExistingCollection(new OpenAIEmbeddings({ openAIApiKey: process.env.OPENAI_API_KEY }), {
+        client: qdrantClient,
+        collectionName: process.env.QDRANT_COLLECTION_NAME,
+      });
+
       const similarDocs = await vectorStore
         .similaritySearch(recentChatHistory, 3, { fileName: companionFileName })
         .catch((err) => {

src/scripts/indexQdrant.mjs

@@ -0,0 +1,48 @@
+// Major ref: https://js.langchain.com/docs/modules/indexes/vector_stores/integrations/qdrant
+import dotenv from "dotenv";
+import { Document } from "langchain/document";
+import { OpenAIEmbeddings } from "langchain/embeddings/openai";
+import { CharacterTextSplitter } from "langchain/text_splitter";
+import { QdrantVectorStore } from "langchain/vectorstores/qdrant";
+import { QdrantClient } from '@qdrant/js-client-rest';
+import fs from "fs";
+import path from "path";
+
+dotenv.config({ path: `.env.local` });
+
+const fileNames = fs.readdirSync("companions");
+const splitter = new CharacterTextSplitter({
+  separator: " ",
+  chunkSize: 200,
+  chunkOverlap: 50, //TODO: adjust both chunk size and chunk overlap later
+});
+
+const langchainDocs = await Promise.all(
+  fileNames.map(async (fileName) => {
+    if (fileName.endsWith(".txt")) {
+      const filePath = path.join("companions", fileName);
+      const fileContent = fs.readFileSync(filePath, "utf8");
+      // get the last section in the doc for background info
+      const lastSection = fileContent.split("###ENDSEEDCHAT###").slice(-1)[0];
+      const splitDocs = await splitter.createDocuments([lastSection]);
+      return splitDocs.map((doc) => {
+        return new Document({
+          metadata: { fileName },
+          pageContent: doc.pageContent,
+        });
+      });
+    }
+  })
+);
+
+
+const qdrantClient = new QdrantClient({ url: process.env.QDRANT_URL, apiKey: process.env?.QDRANT_API_KEY });
+
+await QdrantVectorStore.fromDocuments(
+  langchainDocs.flat().filter((doc) => doc !== undefined),
+  new OpenAIEmbeddings({ openAIApiKey: process.env.OPENAI_API_KEY }),
+  {
+    client: qdrantClient,
+    collectionName: process.env.QDRANT_COLLECTION_NAME,
+  }
+);