diff --git a/frontend/src/components/WorkspaceChat/ChatContainer/index.jsx b/frontend/src/components/WorkspaceChat/ChatContainer/index.jsx
index 6dd1cdf50..372c79a7c 100644
--- a/frontend/src/components/WorkspaceChat/ChatContainer/index.jsx
+++ b/frontend/src/components/WorkspaceChat/ChatContainer/index.jsx
@@ -71,21 +71,6 @@ export default function ChatContainer({ workspace, knownHistory = [] }) {
return false;
}
- // TODO: Delete this snippet once we have streaming stable.
- // const chatResult = await Workspace.sendChat(
- // workspace,
- // promptMessage.userMessage,
- // window.localStorage.getItem(`workspace_chat_mode_${workspace.slug}`) ??
- // "chat",
- // )
- // handleChat(
- // chatResult,
- // setLoadingResponse,
- // setChatHistory,
- // remHistory,
- // _chatHistory
- // )
-
await Workspace.streamChat(
workspace,
promptMessage.userMessage,
diff --git a/frontend/src/models/workspace.js b/frontend/src/models/workspace.js
index aa0b9f744..d015918d4 100644
--- a/frontend/src/models/workspace.js
+++ b/frontend/src/models/workspace.js
@@ -168,22 +168,6 @@ const Workspace = {
const data = await response.json();
return { response, data };
},
-
- // TODO: Deprecated and should be removed from frontend.
- sendChat: async function ({ slug }, message, mode = "query") {
- const chatResult = await fetch(`${API_BASE}/workspace/${slug}/chat`, {
- method: "POST",
- body: JSON.stringify({ message, mode }),
- headers: baseHeaders(),
- })
- .then((res) => res.json())
- .catch((e) => {
- console.error(e);
- return null;
- });
-
- return chatResult;
- },
};
export default Workspace;
diff --git a/server/endpoints/api/workspace/index.js b/server/endpoints/api/workspace/index.js
index 032fe41c3..ffead3adb 100644
--- a/server/endpoints/api/workspace/index.js
+++ b/server/endpoints/api/workspace/index.js
@@ -196,10 +196,11 @@ function apiWorkspaceEndpoints(app) {
return;
}
- await WorkspaceChats.delete({ workspaceId: Number(workspace.id) });
- await DocumentVectors.deleteForWorkspace(Number(workspace.id));
- await Document.delete({ workspaceId: Number(workspace.id) });
- await Workspace.delete({ id: Number(workspace.id) });
+ const workspaceId = Number(workspace.id);
+ await WorkspaceChats.delete({ workspaceId: workspaceId });
+ await DocumentVectors.deleteForWorkspace(workspaceId);
+ await Document.delete({ workspaceId: workspaceId });
+ await Workspace.delete({ id: workspaceId });
try {
await VectorDb["delete-namespace"]({ namespace: slug });
} catch (e) {
@@ -441,7 +442,7 @@ function apiWorkspaceEndpoints(app) {
#swagger.tags = ['Workspaces']
#swagger.description = 'Execute a chat with a workspace'
#swagger.requestBody = {
- description: 'prompt to send to the workspace and the type of conversation (query or chat).',
+ description: 'Send a prompt to the workspace and the type of conversation (query or chat).<br/><b>Query:</b> Will not use LLM unless there are relevant sources from vectorDB & does not recall chat history.<br/><b>Chat:</b> Uses LLM general knowledge w/custom embeddings to produce output, uses rolling chat history.',
required: true,
type: 'object',
content: {
diff --git a/server/endpoints/chat.js b/server/endpoints/chat.js
index d0a2923c5..79fc10132 100644
--- a/server/endpoints/chat.js
+++ b/server/endpoints/chat.js
@@ -1,7 +1,6 @@
const { v4: uuidv4 } = require("uuid");
const { reqBody, userFromSession, multiUserMode } = require("../utils/http");
const { Workspace } = require("../models/workspace");
-const { chatWithWorkspace } = require("../utils/chats");
const { validatedRequest } = require("../utils/middleware/validatedRequest");
const { WorkspaceChats } = require("../models/workspaceChats");
const { SystemSettings } = require("../models/systemSettings");
@@ -95,85 +94,6 @@ function chatEndpoints(app) {
}
}
);
-
- app.post(
- "/workspace/:slug/chat",
- [validatedRequest],
- async (request, response) => {
- try {
- const user = await userFromSession(request, response);
- const { slug } = request.params;
- const { message, mode = "query" } = reqBody(request);
-
- const workspace = multiUserMode(response)
- ? await Workspace.getWithUser(user, { slug })
- : await Workspace.get({ slug });
-
- if (!workspace) {
- response.sendStatus(400).end();
- return;
- }
-
- if (multiUserMode(response) && user.role !== "admin") {
- const limitMessagesSetting = await SystemSettings.get({
- label: "limit_user_messages",
- });
- const limitMessages = limitMessagesSetting?.value === "true";
-
- if (limitMessages) {
- const messageLimitSetting = await SystemSettings.get({
- label: "message_limit",
- });
- const systemLimit = Number(messageLimitSetting?.value);
-
- if (!!systemLimit) {
- const currentChatCount = await WorkspaceChats.count({
- user_id: user.id,
- createdAt: {
- gte: new Date(new Date() - 24 * 60 * 60 * 1000),
- },
- });
-
- if (currentChatCount >= systemLimit) {
- response.status(500).json({
- id: uuidv4(),
- type: "abort",
- textResponse: null,
- sources: [],
- close: true,
- error: `You have met your maximum 24 hour chat quota of ${systemLimit} chats set by the instance administrators. Try again later.`,
- });
- return;
- }
- }
- }
- }
-
- const result = await chatWithWorkspace(workspace, message, mode, user);
- await Telemetry.sendTelemetry(
- "sent_chat",
- {
- multiUserMode: multiUserMode(response),
- LLMSelection: process.env.LLM_PROVIDER || "openai",
- Embedder: process.env.EMBEDDING_ENGINE || "inherit",
- VectorDbSelection: process.env.VECTOR_DB || "pinecone",
- },
- user?.id
- );
- response.status(200).json({ ...result });
- } catch (e) {
- console.error(e);
- response.status(500).json({
- id: uuidv4(),
- type: "abort",
- textResponse: null,
- sources: [],
- close: true,
- error: e.message,
- });
- }
- }
- );
}
module.exports = { chatEndpoints };
diff --git a/server/swagger/openapi.json b/server/swagger/openapi.json
index 184723ed7..7b675c44b 100644
--- a/server/swagger/openapi.json
+++ b/server/swagger/openapi.json
@@ -1598,7 +1598,7 @@
}
},
"requestBody": {
- "description": "prompt to send to the workspace and the type of conversation (query or chat).",
+ "description": "Send a prompt to the workspace and the type of conversation (query or chat).<br/><b>Query:</b> Will not use LLM unless there are relevant sources from vectorDB & does not recall chat history.<br/><b>Chat:</b> Uses LLM general knowledge w/custom embeddings to produce output, uses rolling chat history.",
"required": true,
"type": "object",
"content": {
diff --git a/server/utils/chats/index.js b/server/utils/chats/index.js
index 7e9be6e5b..7fdb47344 100644
--- a/server/utils/chats/index.js
+++ b/server/utils/chats/index.js
@@ -91,6 +91,18 @@ async function chatWithWorkspace(
const hasVectorizedSpace = await VectorDb.hasNamespace(workspace.slug);
const embeddingsCount = await VectorDb.namespaceCount(workspace.slug);
if (!hasVectorizedSpace || embeddingsCount === 0) {
+ if (chatMode === "query") {
+ return {
+ id: uuid,
+ type: "textResponse",
+ sources: [],
+ close: true,
+ error: null,
+ textResponse:
+ "There is no relevant information in this workspace to answer your query.",
+ };
+ }
+
// If there are no embeddings - chat like a normal LLM chat interface.
return await emptyEmbeddingChat({
uuid,
@@ -131,6 +143,20 @@ async function chatWithWorkspace(
};
}
+ // If in query mode and no sources are found, do not
+ // let the LLM try to hallucinate a response or use general knowledge
+ if (chatMode === "query" && sources.length === 0) {
+ return {
+ id: uuid,
+ type: "textResponse",
+ sources: [],
+ close: true,
+ error: null,
+ textResponse:
+ "There is no relevant information in this workspace to answer your query.",
+ };
+ }
+
// Compress message to ensure prompt passes token limit with room for response
// and build system messages based on inputs and history.
const messages = await LLMConnector.compressMessages(
diff --git a/server/utils/chats/stream.js b/server/utils/chats/stream.js
index 73437eec5..11d4effd7 100644
--- a/server/utils/chats/stream.js
+++ b/server/utils/chats/stream.js
@@ -50,6 +50,19 @@ async function streamChatWithWorkspace(
const hasVectorizedSpace = await VectorDb.hasNamespace(workspace.slug);
const embeddingsCount = await VectorDb.namespaceCount(workspace.slug);
if (!hasVectorizedSpace || embeddingsCount === 0) {
+ if (chatMode === "query") {
+ writeResponseChunk(response, {
+ id: uuid,
+ type: "textResponse",
+ textResponse:
+ "There is no relevant information in this workspace to answer your query.",
+ sources: [],
+ close: true,
+ error: null,
+ });
+ return;
+ }
+
// If there are no embeddings - chat like a normal LLM chat interface.
return await streamEmptyEmbeddingChat({
response,
@@ -93,6 +106,21 @@ async function streamChatWithWorkspace(
return;
}
+ // If in query mode and no sources are found, do not
+ // let the LLM try to hallucinate a response or use general knowledge
+ if (chatMode === "query" && sources.length === 0) {
+ writeResponseChunk(response, {
+ id: uuid,
+ type: "textResponse",
+ textResponse:
+ "There is no relevant information in this workspace to answer your query.",
+ sources: [],
+ close: true,
+ error: null,
+ });
+ return;
+ }
+
// Compress message to ensure prompt passes token limit with room for response
// and build system messages based on inputs and history.
const messages = await LLMConnector.compressMessages(