feat(app): add file uploads

Author: ItzCrazyKns

.prettierignore

@@ -35,4 +35,7 @@ coverage
 *.swp
 
 # Ignore all files with the .DS_Store extension (macOS specific)
-.DS_Store
+.DS_Store
+
+# Ignore all files in uploads directory
+uploads

backend.dockerfile

@@ -9,6 +9,7 @@ COPY package.json /home/perplexica/
 COPY yarn.lock /home/perplexica/
 
 RUN mkdir /home/perplexica/data
+RUN mkdir /home/perplexica/uploads
 
 RUN yarn install --frozen-lockfile --network-timeout 600000
 RUN yarn build

docker-compose.yaml

@@ -22,6 +22,7 @@ services:
       - 3001:3001
     volumes:
       - backend-dbstore:/home/perplexica/data
+      - uploads:/home/perplexica/uploads
       - ./config.toml:/home/perplexica/config.toml
     extra_hosts:
       - 'host.docker.internal:host-gateway'
@@ -50,3 +51,4 @@ networks:
 
 volumes:
   backend-dbstore:
+  uploads:

package.json

@@ -6,7 +6,7 @@
   "scripts": {
     "start": "npm run db:push && node dist/app.js",
     "build": "tsc",
-    "dev": "nodemon src/app.ts",
+    "dev": "nodemon --ignore uploads/ src/app.ts ",
     "db:push": "drizzle-kit push sqlite",
     "format": "prettier . --check",
     "format:write": "prettier . --write"
@@ -16,6 +16,7 @@
     "@types/cors": "^2.8.17",
     "@types/express": "^4.17.21",
     "@types/html-to-text": "^9.0.4",
+    "@types/multer": "^1.4.12",
     "@types/pdf-parse": "^1.1.4",
     "@types/readable-stream": "^4.0.11",
     "@types/ws": "^8.5.12",
@@ -41,6 +42,8 @@
     "express": "^4.19.2",
     "html-to-text": "^9.0.5",
     "langchain": "^0.1.30",
+    "mammoth": "^1.8.0",
+    "multer": "^1.4.5-lts.1",
     "pdf-parse": "^1.1.1",
     "winston": "^3.13.0",
     "ws": "^8.17.1",

src/agents/webSearchAgent.ts

@@ -20,10 +20,12 @@ import eventEmitter from 'events';
 import computeSimilarity from '../utils/computeSimilarity';
 import logger from '../utils/logger';
 import LineListOutputParser from '../lib/outputParsers/listLineOutputParser';
-import { getDocumentsFromLinks } from '../lib/linkDocument';
 import LineOutputParser from '../lib/outputParsers/lineOutputParser';
 import { IterableReadableStream } from '@langchain/core/utils/stream';
 import { ChatOpenAI } from '@langchain/openai';
+import path from 'path';
+import fs from 'fs';
+import { getDocumentsFromLinks } from '../utils/documents';
 
 const basicSearchRetrieverPrompt = `
 You are an AI question rephraser. You will be given a conversation and a follow-up question,  you will have to rephrase the follow up question so it is a standalone question and can be used by another LLM to search the web for information to answer it.
@@ -316,6 +318,7 @@ const createBasicWebSearchAnsweringChain = (
   llm: BaseChatModel,
   embeddings: Embeddings,
   optimizationMode: 'speed' | 'balanced' | 'quality',
+  fileIds: string[],
 ) => {
   const basicWebSearchRetrieverChain = createBasicWebSearchRetrieverChain(llm);
 
@@ -336,16 +339,76 @@ const createBasicWebSearchAnsweringChain = (
       return docs;
     }
 
+    const filesData = fileIds
+      .map((file) => {
+        const filePath = path.join(process.cwd(), 'uploads', file);
+
+        const contentPath = filePath + '-extracted.json';
+        const embeddingsPath = filePath + '-embeddings.json';
+
+        const content = JSON.parse(fs.readFileSync(contentPath, 'utf8'));
+        const embeddings = JSON.parse(fs.readFileSync(embeddingsPath, 'utf8'));
+
+        const fileSimilaritySearchObject = content.contents.map(
+          (c: string, i) => {
+            return {
+              fileName: content.title,
+              content: c,
+              embeddings: embeddings.embeddings[i],
+            };
+          },
+        );
+
+        return fileSimilaritySearchObject;
+      })
+      .flat();
+
     if (query.toLocaleLowerCase() === 'summarize') {
-      return docs.slice(0, 15)
+      return docs.slice(0, 15);
     }
 
     const docsWithContent = docs.filter(
       (doc) => doc.pageContent && doc.pageContent.length > 0,
     );
 
     if (optimizationMode === 'speed') {
-      return docsWithContent.slice(0, 15);
+      if (filesData.length > 0) {
+        const [queryEmbedding] = await Promise.all([
+          embeddings.embedQuery(query),
+        ]);
+
+        const fileDocs = filesData.map((fileData) => {
+          return new Document({
+            pageContent: fileData.content,
+            metadata: {
+              title: fileData.fileName,
+              url: `File`,
+            },
+          });
+        });
+
+        const similarity = filesData.map((fileData, i) => {
+          const sim = computeSimilarity(queryEmbedding, fileData.embeddings);
+
+          return {
+            index: i,
+            similarity: sim,
+          };
+        });
+
+        const sortedDocs = similarity
+          .filter((sim) => sim.similarity > 0.3)
+          .sort((a, b) => b.similarity - a.similarity)
+          .slice(0, 8)
+          .map((sim) => fileDocs[sim.index]);
+
+        return [
+          ...sortedDocs,
+          ...docsWithContent.slice(0, 15 - sortedDocs.length),
+        ];
+      } else {
+        return docsWithContent.slice(0, 15);
+      }
     } else if (optimizationMode === 'balanced') {
       const [docEmbeddings, queryEmbedding] = await Promise.all([
         embeddings.embedDocuments(
@@ -354,6 +417,20 @@ const createBasicWebSearchAnsweringChain = (
         embeddings.embedQuery(query),
       ]);
 
+      docsWithContent.push(
+        ...filesData.map((fileData) => {
+          return new Document({
+            pageContent: fileData.content,
+            metadata: {
+              title: fileData.fileName,
+              url: `File`,
+            },
+          });
+        }),
+      );
+
+      docEmbeddings.push(...filesData.map((fileData) => fileData.embeddings));
+
       const similarity = docEmbeddings.map((docEmbedding, i) => {
         const sim = computeSimilarity(queryEmbedding, docEmbedding);
 
@@ -408,6 +485,7 @@ const basicWebSearch = (
   llm: BaseChatModel,
   embeddings: Embeddings,
   optimizationMode: 'speed' | 'balanced' | 'quality',
+  fileIds: string[],
 ) => {
   const emitter = new eventEmitter();
 
@@ -416,6 +494,7 @@ const basicWebSearch = (
       llm,
       embeddings,
       optimizationMode,
+      fileIds,
     );
 
     const stream = basicWebSearchAnsweringChain.streamEvents(
@@ -446,13 +525,15 @@ const handleWebSearch = (
   llm: BaseChatModel,
   embeddings: Embeddings,
   optimizationMode: 'speed' | 'balanced' | 'quality',
+  fileIds: string[],
 ) => {
   const emitter = basicWebSearch(
     message,
     history,
     llm,
     embeddings,
     optimizationMode,
+    fileIds,
   );
   return emitter;
 };

src/db/schema.ts

@@ -1,3 +1,4 @@
+import { sql } from 'drizzle-orm';
 import { text, integer, sqliteTable } from 'drizzle-orm/sqlite-core';
 
 export const messages = sqliteTable('messages', {
@@ -11,9 +12,17 @@ export const messages = sqliteTable('messages', {
   }),
 });
 
+interface File {
+  name: string;
+  fileId: string;
+}
+
 export const chats = sqliteTable('chats', {
   id: text('id').primaryKey(),
   title: text('title').notNull(),
   createdAt: text('createdAt').notNull(),
   focusMode: text('focusMode').notNull(),
+  files: text('files', { mode: 'json' })
+    .$type<File[]>()
+    .default(sql`'[]'`),
 });

src/lib/providers/ollama.ts

@@ -6,7 +6,7 @@ import { ChatOllama } from '@langchain/community/chat_models/ollama';
 export const loadOllamaChatModels = async () => {
   const ollamaEndpoint = getOllamaApiEndpoint();
   const keepAlive = getKeepAlive();
-  
+
   if (!ollamaEndpoint) return {};
 
   try {
@@ -25,7 +25,7 @@ export const loadOllamaChatModels = async () => {
           baseUrl: ollamaEndpoint,
           model: model.model,
           temperature: 0.7,
-          keepAlive: keepAlive
+          keepAlive: keepAlive,
         }),
       };
 

src/routes/index.ts

@@ -7,6 +7,7 @@ import suggestionsRouter from './suggestions';
 import chatsRouter from './chats';
 import searchRouter from './search';
 import discoverRouter from './discover';
+import uploadsRouter from './uploads';
 
 const router = express.Router();
 
@@ -18,5 +19,6 @@ router.use('/suggestions', suggestionsRouter);
 router.use('/chats', chatsRouter);
 router.use('/search', searchRouter);
 router.use('/discover', discoverRouter);
+router.use('/uploads', uploadsRouter);
 
 export default router;