diff --git a/app/constant.ts b/app/constant.ts
index 59a8c1b5048a..b91b931c685a 100644
--- a/app/constant.ts
+++ b/app/constant.ts
@@ -54,3 +54,8 @@ export const OpenaiPath = {
 };
 
 export const DEFAULT_INPUT_TEMPLATE = `{{input}}`; // input / time / model / lang
+export const DEFAULT_SYSTEM_TEMPLATE = `
+You are ChatGPT, a large language model trained by OpenAI.
+Knowledge cutoff: 2021-09
+Current model: {{model}}
+Current time: {{time}}`;
diff --git a/app/store/chat.ts b/app/store/chat.ts
index fa6296811bfe..4c466a295c56 100644
--- a/app/store/chat.ts
+++ b/app/store/chat.ts
@@ -7,7 +7,11 @@ import Locale, { getLang } from "../locales";
 import { showToast } from "../components/ui-lib";
 import { ModelConfig, ModelType, useAppConfig } from "./config";
 import { createEmptyMask, Mask } from "./mask";
-import { DEFAULT_INPUT_TEMPLATE, StoreKey } from "../constant";
+import {
+  DEFAULT_INPUT_TEMPLATE,
+  DEFAULT_SYSTEM_TEMPLATE,
+  StoreKey,
+} from "../constant";
 import { api, RequestMessage } from "../client/api";
 import { ChatControllerPool } from "../client/controller";
 import { prettyObject } from "../utils/format";
@@ -279,7 +283,7 @@ export const useChatStore = create<ChatStore>()(
         const modelConfig = session.mask.modelConfig;
 
         const userContent = fillTemplateWith(content, modelConfig);
-        console.log("[User Input] fill with template: ", userContent);
+        console.log("[User Input] after template: ", userContent);
 
         const userMessage: ChatMessage = createMessage({
           role: "user",
@@ -312,7 +316,6 @@ export const useChatStore = create<ChatStore>()(
         });
 
         // make request
-        console.log("[User Input] ", sendMessages);
         api.llm.chat({
           messages: sendMessages,
           config: { ...modelConfig, stream: true },
@@ -391,6 +394,27 @@ export const useChatStore = create<ChatStore>()(
         // in-context prompts
         const contextPrompts = session.mask.context.slice();
 
+        // system prompts, to get close to OpenAI Web ChatGPT
+        // only will be injected if user does not use a mask or set none context prompts
+        const shouldInjectSystemPrompts = contextPrompts.length === 0;
+        const systemPrompts = shouldInjectSystemPrompts
+          ? [
+              createMessage({
+                role: "system",
+                content: fillTemplateWith("", {
+                  ...modelConfig,
+                  template: DEFAULT_SYSTEM_TEMPLATE,
+                }),
+              }),
+            ]
+          : [];
+        if (shouldInjectSystemPrompts) {
+          console.log(
+            "[Global System Prompt] ",
+            systemPrompts.at(0)?.content ?? "empty",
+          );
+        }
+
         // long term memory
         const shouldSendLongTermMemory =
           modelConfig.sendMemory &&
@@ -409,6 +433,7 @@ export const useChatStore = create<ChatStore>()(
         );
 
         // lets concat send messages, including 4 parts:
+        // 0. system prompt: to get close to OpenAI Web ChatGPT
         // 1. long term memory: summarized memory messages
         // 2. pre-defined in-context prompts
         // 3. short term memory: latest n messages
@@ -435,6 +460,7 @@ export const useChatStore = create<ChatStore>()(
 
         // concat all messages
         const recentMessages = [
+          ...systemPrompts,
           ...longTermMemoryPrompts,
           ...contextPrompts,
           ...reversedRecentMessages.reverse(),
diff --git a/app/store/config.ts b/app/store/config.ts
index b15fa9148021..945e1be7c45d 100644
--- a/app/store/config.ts
+++ b/app/store/config.ts
@@ -177,9 +177,9 @@ export const useAppConfig = create<ChatConfigStore>()(
     }),
     {
       name: StoreKey.Config,
-      version: 3.1,
+      version: 3.2,
       migrate(persistedState, version) {
-        if (version === 3.1) return persistedState as any;
+        if (version === 3.2) return persistedState as any;
 
         const state = persistedState as ChatConfig;
         state.modelConfig.sendMemory = true;