feat: add voyage embeddings (#171)

Closes #169

README.md

@@ -13,12 +13,12 @@
 
 </div>
 
-Rocketnotes is a web-based Markdown note taking app with LLM-powered text completion, chat and semantic search.  
+Rocketnotes is a web-based Markdown note taking app with LLM-powered text completion, chat and semantic search.
 It utilizes a [100% Serverless RAG pipeline](https://medium.com/@fynnfluegge/serverless-rag-on-aws-bf8029f8bffd) built with
 [langchain](https://github.com/langchain-ai/langchain),
 [sentence-transformers](https://github.com/UKPLab/sentence-transformers),
 [faiss](https://github.com/facebookresearch/faiss),
-[Ollama](https://github.com/jmorganca/ollama) and OpenAI or Anthropic.
+[Ollama](https://github.com/jmorganca/ollama), OpenAI, Anthropic and Voyage.
 
 ## How to use
 
@@ -39,7 +39,7 @@ It utilizes a [100% Serverless RAG pipeline](https://medium.com/@fynnfluegge/ser
 - 🔦 Semantic search
 - ✍️ Copilot-like text completion
 - 🤖 Chat with your documents
-  - Serverless RAG with faiss, OpenAI and/or Anthropic
+  - Serverless RAG with faiss, OpenAI, Anthropic, Voyage
 - 📦 Local mode with Docker
   - use Ollama and/or Sentence Transformers for 100% local RAG
 - 📥 Zettelkasten with semantic archiving

lambda-handler/chat-handler/main.py

@@ -6,7 +6,11 @@
 from langchain.chains import ConversationalRetrievalChain
 from langchain.memory import ConversationSummaryMemory
 from langchain_anthropic import ChatAnthropic
-from langchain_community.embeddings import HuggingFaceEmbeddings, OllamaEmbeddings
+from langchain_community.embeddings import (
+    HuggingFaceEmbeddings,
+    OllamaEmbeddings,
+    VoyageEmbeddings,
+)
 from langchain_community.llms import Ollama
 from langchain_community.vectorstores import FAISS
 from langchain_openai import ChatOpenAI, OpenAIEmbeddings
@@ -72,6 +76,12 @@ def handler(event, context):
         else:
             return {"statusCode": 400, "body": "OpenAI API key is missing"}
 
+    if embeddings_model == "voyage-2":
+        if "voyageApiKey" in userConfig:
+            os.environ["VOYAGE_API_KEY"] = userConfig.get("voyageApiKey").get("S")
+        else:
+            return {"statusCode": 400, "body": "OpenAI API key is missing"}
+
     if "llm" in userConfig:
         llm_model = userConfig.get("llm").get("S")
     else:
@@ -106,6 +116,8 @@ def handler(event, context):
 
     if embeddings_model == "text-embedding-ada-002":
         embeddings = OpenAIEmbeddings(client=None, model="text-embedding-ada-002")
+    elif embeddings_model == "voyage-2":
+        embeddings = VoyageEmbeddings(model=embeddings_model)
     elif embeddings_model == "Sentence-Transformers":
         embeddings = HuggingFaceEmbeddings(model_kwargs={"device": "cpu"})
     elif embeddings_model == "Ollama-nomic-embed-text":

lambda-handler/get-user-config-handler/main.go

@@ -21,6 +21,7 @@ type UserConfig struct {
 	Llm string `json:"llm"`
 	OpenAiApiKey string `json:"openAiApiKey"`
 	AnthropicApiKey string `json:"anthropicApiKey"`
+	VoyageApiKey string `json:"voyageApiKey"`
 }
 
 func init() {

lambda-handler/save-user-config-handler/main.go

@@ -21,6 +21,7 @@ type RequestBody struct {
 	Llm string `json:"llm"`
 	OpenAiApiKey string `json:"openAiApiKey"`
 	AnthropicApiKey string `json:"anthropicApiKey"`
+	VoyageApiKey string `json:"voyageApiKey`
 	RecreateIndex bool `json:"recreateIndex"`
 }
 
@@ -30,12 +31,14 @@ type UserConfig struct {
 	Llm string `json:"llm"`
 	OpenAiApiKey string `json:"openAiApiKey"`
 	AnthropicApiKey string `json:"anthropicApiKey"`
+	VoyageApiKey string `json:"voyageApiKey"`
 }
 
 type SqsMessage struct {
   UserId string `json:"userId"`
   OpenAiApiKey string `json:"openAiApiKey"`
   AnthropicApiKey string `json:"anthropicApiKey"`
+	VoyageApiKey string `json:"voyageApiKey`
   RecreateIndex bool `json:"recreateIndex"`
 }
 
@@ -63,7 +66,7 @@ func handleRequest(ctx context.Context, request events.APIGatewayProxyRequest) (
 
 	tableName := "tnn-UserConfig"
 
-	av, err := dynamodbattribute.MarshalMap(UserConfig{item.Id, item.EmbeddingModel, item.Llm, item.OpenAiApiKey, item.AnthropicApiKey})
+	av, err := dynamodbattribute.MarshalMap(UserConfig{item.Id, item.EmbeddingModel, item.Llm, item.OpenAiApiKey, item.AnthropicApiKey, item.VoyageApiKey})
 
 	if err != nil {
 		log.Fatalf("Got error marshalling new document item: %s", err)
@@ -83,7 +86,7 @@ func handleRequest(ctx context.Context, request events.APIGatewayProxyRequest) (
 	if os.Getenv("USE_LOCAL_DYNAMODB") != "1" && item.RecreateIndex {
 		qsvc := sqs.New(sess)
 
-		m := SqsMessage{item.Id, item.OpenAiApiKey, item.AnthropicApiKey, item.RecreateIndex}
+		m := SqsMessage{item.Id, item.OpenAiApiKey, item.AnthropicApiKey, item.VoyageApiKey, item.RecreateIndex}
 		b, err := json.Marshal(m)
 
 		_, err = qsvc.SendMessage(&sqs.SendMessageInput{

lambda-handler/save-vector-embeddings-handler/main.py

@@ -8,7 +8,11 @@
     MarkdownHeaderTextSplitter,
     RecursiveCharacterTextSplitter,
 )
-from langchain_community.embeddings import HuggingFaceEmbeddings, OllamaEmbeddings
+from langchain_community.embeddings import (
+    HuggingFaceEmbeddings,
+    OllamaEmbeddings,
+    VoyageEmbeddings,
+)
 from langchain_community.vectorstores import FAISS
 from langchain_openai import OpenAIEmbeddings
 
@@ -53,11 +57,11 @@ def handler(event, context):
     userConfig = userConfig["Item"]
     embeddingsModel = userConfig.get("embeddingModel", {}).get("S", None)
     openAiApiKey = userConfig.get("openAiApiKey", {}).get("S", None)
-    anthropicApiKey = userConfig.get("anthropicApiKey", {}).get("S", None)
+    voyageApiKey = userConfig.get("voyageApiKey", {}).get("S", None)
     if openAiApiKey is not None:
         os.environ["OPENAI_API_KEY"] = openAiApiKey
-    if anthropicApiKey is not None:
-        os.environ["ANTHROPIC_API_KEY"] = anthropicApiKey
+    if voyageApiKey is not None:
+        os.environ["VOYAGE_API_KEY"] = voyageApiKey
 
     if embeddingsModel == "text-embedding-ada-002":
         if openAiApiKey is None:
@@ -66,6 +70,13 @@ def handler(event, context):
                 "body": json.dumps("OpenAI API key not found"),
             }
         embeddings = OpenAIEmbeddings(client=None, model=embeddingsModel)
+    elif embeddingsModel == "voyage-2":
+        if voyageApiKey is None:
+            return {
+                "statusCode": 400,
+                "body": json.dumps("Voyage API key not found"),
+            }
+        embeddings = VoyageEmbeddings(model=embeddingsModel)
     elif embeddingsModel == "Sentence-Transformers":
         embeddings = HuggingFaceEmbeddings(model_kwargs={"device": "cpu"})
     elif embeddingsModel == "Ollama-nomic-embed-text":
@@ -140,10 +151,10 @@ def handler(event, context):
         else:
             # Faiss index does not exist, create the index from scratch
             # ---------------------------------------------------------
-            filter_expression = "userId = :user_value AND deleted = :deleted_value"
+            filter_expression = "userId = :user_value"
             expression_attribute_values = {
                 ":user_value": {"S": userId},
-                ":deleted_value": {"BOOL": False},
+                # ":deleted_value": {"BOOL": False},
             }
 
             # Execute the query
@@ -159,6 +170,8 @@ def handler(event, context):
             if documents:
                 split_documents = []
                 for document in documents:
+                    if document["deleted"]["BOOL"]:
+                        continue
                     try:
                         content = document["content"]["S"]
                         documentId = document["id"]["S"]

lambda-handler/semantic-search-handler/main.py

@@ -3,7 +3,11 @@
 from pathlib import Path
 
 import boto3
-from langchain_community.embeddings import HuggingFaceEmbeddings, OllamaEmbeddings
+from langchain_community.embeddings import (
+    HuggingFaceEmbeddings,
+    OllamaEmbeddings,
+    VoyageEmbeddings,
+)
 from langchain_community.vectorstores import FAISS
 from langchain_openai import OpenAIEmbeddings
 
@@ -64,7 +68,13 @@ def handler(event, context):
             os.environ["OPENAI_API_KEY"] = userConfig.get("openAiApiKey").get("S")
         else:
             return {"statusCode": 400, "body": "OpenAI API key is missing"}
-        embeddings = OpenAIEmbeddings(client=None, model="text-embedding-ada-002")
+        embeddings = OpenAIEmbeddings(client=None, model=embeddings_model)
+    elif embeddings_model == "voyage-2":
+        if "voyageApiKey" in userConfig:
+            os.environ["VOYAGE_API_KEY"] = userConfig.get("voyageApiKey").get("S")
+        else:
+            return {"statusCode": 400, "body": "OpenAI API key is missing"}
+        embeddings = VoyageEmbeddings(model=embeddings_model)
     elif embeddings_model == "Sentence-Transformers":
         embeddings = HuggingFaceEmbeddings(model_kwargs={"device": "cpu"})
     elif embeddings_model == "Ollama-nomic-embed-text":

landing-page/src/app/app.component.html

@@ -107,7 +107,7 @@ <h3 class="h3 mb-3">
               >Ollama</a
             >
             <br />
-            and OpenAI or Anthropic.
+            and OpenAI/Anthropic/Voyage
           </p>
           <p>
             Read details below to learn more about all feature of the powerful

webapp/src/app/component/dialog-config/config-dialog.component.html

@@ -10,6 +10,14 @@ <h5>Embeddings Model</h5>
           value="text-embedding-ada-002"
         />Text-Embedding-Ada-002
       </label>
+      <label class="radio-inline">
+        <input
+          type="radio"
+          name="embeddingModel"
+          [(ngModel)]="selectedEmbeddingModel"
+          value="voyage-2"
+        />Voyage-2
+      </label>
       <label class="radio-inline" *ngIf="isLocal">
         <input
           type="radio"
@@ -158,6 +166,25 @@ <h5>Anthropic API Key</h5>
           key
         </span>
       </div>
+      <h5>Voyage API Key</h5>
+      <textarea
+        type="text"
+        id="voyageApiKey"
+        rows="2"
+        [placeholder]="'Enter your Voyage API key...'"
+        style="width: 94%"
+        [(ngModel)]="voyageApiKey"
+        #message="ngModel"
+        name="voyageApiKey"
+      ></textarea>
+      <div>
+        <span
+          id="voyageApiKeyRequired"
+          style="text-align: center; color: var(--warning-color); display: none"
+        >
+          For the selected configuration, you need to provide a Voyage API key
+        </span>
+      </div>
       <br />
       <div>
         <span style="text-align: right"

webapp/src/app/component/dialog-config/config-dialog.component.ts

@@ -18,6 +18,7 @@ export class ConfigDialogComponent implements OnDestroy, OnInit {
   selectedLlm: string = 'gpt-3.5-turbo';
   openAiApiKey: string;
   anthropicApiKey: string;
+  voyageApiKey: string;
   isLocal: boolean = !environment.production;
 
   constructor(
@@ -42,6 +43,7 @@ export class ConfigDialogComponent implements OnDestroy, OnInit {
             }
             this.openAiApiKey = config['openAiApiKey'] ?? '';
             this.anthropicApiKey = config['anthropicApiKey'] ?? '';
+            this.voyageApiKey = config['voyageApiKey'] ?? '';
           });
         const overlay = document.getElementById('configDialog');
         if (overlay.getAttribute('outsideClickListener') !== 'true') {
@@ -84,13 +86,28 @@ export class ConfigDialogComponent implements OnDestroy, OnInit {
         'anthropicApiKeyRequired',
       );
       anthropicApiKeyRequiredWarning.style.display = 'none';
+
+      if (this.selectedEmbeddingModel === 'voyage-2' && !this.voyageApiKey) {
+        const voyageApiKeyRequiredWarning = document.getElementById(
+          'voyageApiKeyRequired',
+        );
+        voyageApiKeyRequiredWarning.style.display = 'block';
+        return;
+      } else {
+        const voyageApiKeyRequiredWarning = document.getElementById(
+          'voyageApiKeyRequired',
+        );
+        voyageApiKeyRequiredWarning.style.display = 'none';
+      }
+
       this.restService
         .post('userConfig', {
           id: localStorage.getItem('currentUserId'),
           embeddingModel: this.selectedEmbeddingModel,
           llm: this.selectedLlm,
           openAiApiKey: this.openAiApiKey,
           anthropicApiKey: this.anthropicApiKey,
+          voyageApiKey: this.voyageApiKey,
           recreateIndex:
             this.currentEmbeddingModel !== this.selectedEmbeddingModel,
         })
@@ -109,6 +126,7 @@ export class ConfigDialogComponent implements OnDestroy, OnInit {
               llm: this.selectedLlm,
               openAiApiKey: this.openAiApiKey,
               anthropicApiKey: this.anthropicApiKey,
+              voyageApiKey: this.voyageApiKey,
             }),
           );
           if (