Add Custom OpenAI Embedding model, better statuses, system theme, fix copying, other bug fixes

app.dockerfile

@@ -31,5 +31,6 @@ COPY --from=builder /home/perplexica/migrator/index.js ./migrate.js
 RUN mkdir /home/perplexica/uploads
 
 COPY entrypoint.sh ./entrypoint.sh
+RUN sed -i 's/\r$//' ./entrypoint.sh
 RUN chmod +x ./entrypoint.sh
-CMD ["./entrypoint.sh"]
+CMD ["/bin/sh","/home/perplexica/entrypoint.sh"]

src/app/api/chat/route.ts

@@ -61,10 +61,22 @@ const handleEmitterEvents = async (
 ) => {
   let recievedMessage = '';
   let sources: any[] = [];
+  let sentGeneratingStatus = false;
 
-  stream.on('data', (data) => {
+  stream.on('data', (data: string) => {
     const parsedData = JSON.parse(data);
     if (parsedData.type === 'response') {
+      if (!sentGeneratingStatus) {
+        writer.write(
+          encoder.encode(
+            JSON.stringify({
+              type: 'status',
+              data: 'Generating answer...',
+            }) + '\n',
+          ),
+        );
+        sentGeneratingStatus = true;
+      }
       writer.write(
         encoder.encode(
           JSON.stringify({
@@ -77,6 +89,17 @@ const handleEmitterEvents = async (
 
       recievedMessage += parsedData.data;
     } else if (parsedData.type === 'sources') {
+      if (!sentGeneratingStatus) {
+        writer.write(
+          encoder.encode(
+            JSON.stringify({
+              type: 'status',
+              data: 'Generating answer...',
+            }) + '\n',
+          ),
+        );
+        sentGeneratingStatus = true;
+      }
       writer.write(
         encoder.encode(
           JSON.stringify({
@@ -114,8 +137,16 @@ const handleEmitterEvents = async (
       })
       .execute();
   });
-  stream.on('error', (data) => {
+  stream.on('error', (data: string) => {
     const parsedData = JSON.parse(data);
+    writer.write(
+      encoder.encode(
+        JSON.stringify({
+          type: 'status',
+          data: 'Chat completion failed.',
+        }) + '\n',
+      ),
+    );
     writer.write(
       encoder.encode(
         JSON.stringify({
@@ -218,6 +249,28 @@ export const POST = async (req: Request) => {
         body.embeddingModel?.name || Object.keys(embeddingProvider)[0]
       ];
 
+    const selectedChatProviderKey =
+      body.chatModel?.provider || Object.keys(chatModelProviders)[0];
+    const selectedChatModelKey =
+      body.chatModel?.name || Object.keys(chatModelProvider)[0];
+    const selectedEmbeddingProviderKey =
+      body.embeddingModel?.provider || Object.keys(embeddingModelProviders)[0];
+    const selectedEmbeddingModelKey =
+      body.embeddingModel?.name || Object.keys(embeddingProvider)[0];
+
+    console.log('[Models] Chat request', {
+      chatProvider: selectedChatProviderKey,
+      chatModel: selectedChatModelKey,
+      embeddingProvider: selectedEmbeddingProviderKey,
+      embeddingModel: selectedEmbeddingModelKey,
+      ...(selectedChatProviderKey === 'custom_openai'
+        ? { chatBaseURL: getCustomOpenaiApiUrl() }
+        : {}),
+      ...(selectedEmbeddingProviderKey === 'custom_openai'
+        ? { embeddingBaseURL: getCustomOpenaiApiUrl() }
+        : {}),
+    });
+
     let llm: BaseChatModel | undefined;
     let embedding = embeddingModel.model;
 
@@ -272,11 +325,54 @@ export const POST = async (req: Request) => {
       );
     }
 
+    const llmProxy = new Proxy(llm as any, {
+      get(target, prop, receiver) {
+        if (
+          prop === 'invoke' ||
+          prop === 'stream' ||
+          prop === 'streamEvents' ||
+          prop === 'generate'
+        ) {
+          return (...args: any[]) => {
+            console.log('[Models] Chat model call', {
+              provider: selectedChatProviderKey,
+              model: selectedChatModelKey,
+              method: String(prop),
+            });
+            return (target as any)[prop](...args);
+          };
+        }
+        return Reflect.get(target, prop, receiver);
+      },
+    });
+
+    const embeddingProxy = new Proxy(embedding as any, {
+      get(target, prop, receiver) {
+        if (prop === 'embedQuery' || prop === 'embedDocuments') {
+          return (...args: any[]) => {
+            console.log('[Models] Embedding model call', {
+              provider: selectedEmbeddingProviderKey,
+              model: selectedEmbeddingModelKey,
+              method: String(prop),
+              size:
+                prop === 'embedDocuments'
+                  ? Array.isArray(args[0])
+                    ? args[0].length
+                    : undefined
+                  : undefined,
+            });
+            return (target as any)[prop](...args);
+          };
+        }
+        return Reflect.get(target, prop, receiver);
+      },
+    });
+
     const stream = await handler.searchAndAnswer(
       message.content,
       history,
-      llm,
-      embedding,
+      llmProxy as any,
+      embeddingProxy as any,
       body.optimizationMode,
       body.files,
       body.systemInstructions,
@@ -286,6 +382,18 @@ export const POST = async (req: Request) => {
     const writer = responseStream.writable.getWriter();
     const encoder = new TextEncoder();
 
+    writer.write(
+      encoder.encode(
+        JSON.stringify({
+          type: 'status',
+          data:
+            body.focusMode === 'writingAssistant'
+              ? 'Waiting for chat completion...'
+              : 'Searching web...',
+        }) + '\n',
+      ),
+    );
+
     handleEmitterEvents(stream, writer, encoder, aiMessageId, message.chatId);
     handleHistorySave(message, humanMessageId, body.focusMode, body.files);
 

src/app/api/search/route.ts

@@ -75,6 +75,19 @@ export const POST = async (req: Request) => {
       body.embeddingModel?.name ||
       Object.keys(embeddingModelProviders[embeddingModelProvider])[0];
 
+    console.log('[Models] Search request', {
+      chatProvider: chatModelProvider,
+      chatModel,
+      embeddingProvider: embeddingModelProvider,
+      embeddingModel,
+      ...(chatModelProvider === 'custom_openai'
+        ? { chatBaseURL: getCustomOpenaiApiUrl() }
+        : {}),
+      ...(embeddingModelProvider === 'custom_openai'
+        ? { embeddingBaseURL: getCustomOpenaiApiUrl() }
+        : {}),
+    });
+
     let llm: BaseChatModel | undefined;
     let embeddings: Embeddings | undefined;
 
@@ -118,11 +131,54 @@ export const POST = async (req: Request) => {
       return Response.json({ message: 'Invalid focus mode' }, { status: 400 });
     }
 
+    const llmProxy = new Proxy(llm as any, {
+      get(target, prop, receiver) {
+        if (
+          prop === 'invoke' ||
+          prop === 'stream' ||
+          prop === 'streamEvents' ||
+          prop === 'generate'
+        ) {
+          return (...args: any[]) => {
+            console.log('[Models] Chat model call', {
+              provider: chatModelProvider,
+              model: chatModel,
+              method: String(prop),
+            });
+            return (target as any)[prop](...args);
+          };
+        }
+        return Reflect.get(target, prop, receiver);
+      },
+    });
+
+    const embeddingProxy = new Proxy(embeddings as any, {
+      get(target, prop, receiver) {
+        if (prop === 'embedQuery' || prop === 'embedDocuments') {
+          return (...args: any[]) => {
+            console.log('[Models] Embedding model call', {
+              provider: embeddingModelProvider,
+              model: embeddingModel,
+              method: String(prop),
+              size:
+                prop === 'embedDocuments'
+                  ? Array.isArray(args[0])
+                    ? args[0].length
+                    : undefined
+                  : undefined,
+            });
+            return (target as any)[prop](...args);
+          };
+        }
+        return Reflect.get(target, prop, receiver);
+      },
+    });
+
     const emitter = await searchHandler.searchAndAnswer(
       body.query,
       history,
-      llm,
-      embeddings,
+      llmProxy as any,
+      embeddingProxy as any,
       body.optimizationMode,
       [],
       body.systemInstructions || '',

src/app/api/uploads/route.ts

@@ -3,6 +3,7 @@ import fs from 'fs';
 import path from 'path';
 import crypto from 'crypto';
 import { getAvailableEmbeddingModelProviders } from '@/lib/providers';
+import { getCustomOpenaiApiUrl } from '@/lib/config';
 import { PDFLoader } from '@langchain/community/document_loaders/fs/pdf';
 import { DocxLoader } from '@langchain/community/document_loaders/fs/docx';
 import { RecursiveCharacterTextSplitter } from '@langchain/textsplitters';
@@ -46,6 +47,14 @@ export async function POST(req: Request) {
     const embeddingModel =
       embedding_model ?? Object.keys(embeddingModels[provider as string])[0];
 
+    console.log('[Models] Upload embeddings request', {
+      embeddingProvider: provider,
+      embeddingModel,
+      ...(provider === 'custom_openai'
+        ? { embeddingBaseURL: getCustomOpenaiApiUrl() }
+        : {}),
+    });
+
     let embeddingsModel =
       embeddingModels[provider as string]?.[embeddingModel as string]?.model;
     if (!embeddingsModel) {
@@ -55,6 +64,28 @@ export async function POST(req: Request) {
       );
     }
 
+    const loggedEmbeddings = new Proxy(embeddingsModel as any, {
+      get(target, prop, receiver) {
+        if (prop === 'embedQuery' || prop === 'embedDocuments') {
+          return (...args: any[]) => {
+            console.log('[Models] Upload embedding model call', {
+              provider,
+              model: embeddingModel,
+              method: String(prop),
+              size:
+                prop === 'embedDocuments'
+                  ? Array.isArray(args[0])
+                    ? args[0].length
+                    : undefined
+                  : undefined,
+            });
+            return (target as any)[prop](...args);
+          };
+        }
+        return Reflect.get(target, prop, receiver);
+      },
+    });
+
     const processedFiles: FileRes[] = [];
 
     await Promise.all(
@@ -98,7 +129,7 @@ export async function POST(req: Request) {
           }),
         );
 
-        const embeddings = await embeddingsModel.embedDocuments(
+        const embeddings = await loggedEmbeddings.embedDocuments(
           splitted.map((doc) => doc.pageContent),
         );
         const embeddingsDataPath = filePath.replace(

src/components/Chat.tsx

@@ -16,6 +16,7 @@ const Chat = ({
   setFileIds,
   files,
   setFiles,
+  statusText,
 }: {
   messages: Message[];
   sendMessage: (message: string) => void;
@@ -26,6 +27,7 @@ const Chat = ({
   setFileIds: (fileIds: string[]) => void;
   files: File[];
   setFiles: (files: File[]) => void;
+  statusText?: string;
 }) => {
   const [dividerWidth, setDividerWidth] = useState(0);
   const dividerRef = useRef<HTMLDivElement | null>(null);
@@ -78,14 +80,17 @@ const Chat = ({
               isLast={isLast}
               rewrite={rewrite}
               sendMessage={sendMessage}
+              statusText={statusText}
             />
             {!isLast && msg.role === 'assistant' && (
               <div className="h-px w-full bg-light-secondary dark:bg-dark-secondary" />
             )}
           </Fragment>
         );
       })}
-      {loading && !messageAppeared && <MessageBoxLoading />}
+      {loading && !messageAppeared && (
+        <MessageBoxLoading statusText={statusText} />
+      )}
       <div ref={messageEnd} className="h-0" />
       {dividerWidth > 0 && (
         <div