feat: Qdrant vector store

.env.local.example

@@ -21,4 +21,9 @@ PINECONE_INDEX=ai****
 
 # Supabase related environment variables
 SUPABASE_URL=https://****
-SUPABASE_PRIVATE_KEY=eyJ****
+SUPABASE_PRIVATE_KEY=eyJ****
+
+# Qdrant related environment variables
+QDRANT_URL="https://****"
+QDRANT_API_KEY=****
+QDRANT_COLLECTION_NAME=https://****

README.md

@@ -13,7 +13,7 @@
 
 - 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/) 
 - Image Model: [Replicate](https://replicate.com/)
 - Text Model: [OpenAI](https://platform.openai.com/docs/models)
@@ -63,7 +63,7 @@ c. **Replicate API key**
 
 Visit https://replicate.com/account/api-tokens to get your Replicate API key
 
-> **_NOTE:_**  By default, this template uses Pinecone as vector store, but you can turn on Supabase pgvector easily. 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**
 - Create a Pinecone index by visiting https://app.pinecone.io/ and click on "Create Index"
@@ -97,6 +97,13 @@ In `QAModel.tsx`, replace `/api/qa-pinecone` with `/api/qa-pg-vector`. Then run
 npm run generate-embeddings-supabase
 ```
 
+#### If using Qdrant
+
+In `QAModel.tsx`, replace `/api/qa-pinecone` with `/api/qa-qdrant`. Then run the following command to generate embeddings and store them in Qdrant:
+
+```bash
+npm run generate-embeddings-qdrant
+```
 
 ### 5. Run app locally
 

package.json

@@ -8,12 +8,14 @@
     "start": "next start",
     "lint": "next lint",
     "generate-embeddings-pinecone": "node src/scripts/indexBlogs.mjs",
-    "generate-embeddings-supabase": "node src/scripts/indexBlogPGVector.mjs"
+    "generate-embeddings-supabase": "node src/scripts/indexBlogPGVector.mjs",
+    "generate-embeddings-qdrant": "node src/scripts/indexBlogsQdrant.mjs"
   },
   "dependencies": {
     "@arcjet/next": "^1.0.0-alpha.13",
     "@clerk/nextjs": "^5.1.4",
     "@headlessui/react": "^1.7.15",
+    "@qdrant/js-client-rest": "^1.9.0",
     "@pinecone-database/pinecone": "^2.2.2",
     "@supabase/supabase-js": "^2.25.0",
     "@tailwindcss/forms": "^0.5.7",

src/app/api/qa-qdrant/route.ts

@@ -0,0 +1,36 @@
+import dotenv from "dotenv";
+import { VectorDBQAChain } from "langchain/chains";
+import { OpenAIEmbeddings } from "langchain/embeddings/openai";
+import { OpenAI } from "langchain/llms/openai";
+import { StreamingTextResponse, LangChainStream } from "ai";
+import { CallbackManager } from "langchain/callbacks";
+import { QdrantVectorStore } from "langchain/vectorstores/qdrant";
+import { QdrantClient } from '@qdrant/js-client-rest';
+
+dotenv.config({ path: `.env.local` });
+
+export async function POST(request: Request) {
+  const { prompt } = await request.json();
+  const qdrantClient = new QdrantClient({ url: process.env.QDRANT_URL, apiKey: process.env?.QDRANT_API_KEY });
+
+  const vectorStore = await QdrantVectorStore.fromExistingCollection(new OpenAIEmbeddings({ openAIApiKey: process.env.OPENAI_API_KEY }), {
+    client: qdrantClient,
+    collectionName: process.env.QDRANT_COLLECTION_NAME,
+  });
+
+  const { stream, handlers } = LangChainStream();
+  const model = new OpenAI({
+    streaming: true,
+    modelName: "gpt-3.5-turbo-16k",
+    openAIApiKey: process.env.OPENAI_API_KEY,
+    callbackManager: CallbackManager.fromHandlers(handlers),
+  });
+
+  const chain = VectorDBQAChain.fromLLM(model, vectorStore, {
+    k: 1,
+    returnSourceDocuments: true,
+  });
+  chain.call({ query: prompt }).catch(console.error);
+
+  return new StreamingTextResponse(stream);
+}

src/scripts/indexBlogsQdrant.mjs

@@ -0,0 +1,57 @@
+// 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 {QdrantClient} from '@qdrant/js-client-rest';
+import { QdrantVectorStore } from "langchain/vectorstores/qdrant";
+import fs from "fs";
+import path from "path";
+
+dotenv.config({ path: `.env.local` });
+
+const MAX_TOKENS = 8191;
+
+const fileNames = fs.readdirSync("blogs");
+
+// Helper function to chunk a string into smaller parts
+const chunkString = (str, length) => {
+  const size = Math.ceil(str.length / length);
+  const r = Array(size);
+  let offset = 0;
+
+  for (let i = 0; i < size; i++) {
+    r[i] = str.substr(offset, length);
+    offset += length;
+  }
+
+  return r;
+};
+
+const lanchainDocs = [];
+
+fileNames.forEach((fileName) => {
+  const filePath = path.join("blogs", fileName);
+  const fileContent = fs.readFileSync(filePath, "utf8");
+
+  const chunks = chunkString(fileContent, MAX_TOKENS);
+
+  chunks.forEach((chunk, index) => {
+    lanchainDocs.push(new Document({
+      metadata: { fileName, chunkIndex: index },
+      pageContent: chunk,
+    }));
+  });
+});
+
+
+const qdrantClient = new QdrantClient({ url: process.env.QDRANT_URL, apiKey: process.env?.QDRANT_API_KEY });
+
+await QdrantVectorStore.fromDocuments(
+  langchainDocs,
+  new OpenAIEmbeddings({ openAIApiKey: process.env.OPENAI_API_KEY }),
+  {
+    client: qdrantClient,
+    collectionName: process.env.QDRANT_COLLECTION_NAME,
+  }
+);