⚡ (openai) Truncate messages sequence automatically if reaching token…

… limit

apps/builder/src/features/templates/components/TemplatesModal.tsx

@@ -77,6 +77,8 @@ export const TemplatesModal = ({ isOpen, onClose, onTypebotChoose }: Props) => {
             borderRightWidth={1}
             justify="space-between"
             flexShrink={0}
+            overflowY="scroll"
+            className="hide-scrollbar"
           >
             <Stack spacing={5}>
               <Stack spacing={2}>

apps/builder/src/hooks/useToast.tsx

@@ -17,7 +17,7 @@ export const useToast = () => {
     }: Omit<ToastProps, 'onClose'>) => {
       toast({
         position: 'top-right',
-        duration: details ? null : undefined,
+        duration: details && status === 'error' ? null : undefined,
         render: ({ onClose }) => (
           <Toast
             title={title}

apps/viewer/package.json

@@ -13,14 +13,15 @@
     "test:report": "pnpm playwright show-report"
   },
   "dependencies": {
+    "@dqbd/tiktoken": "^1.0.7",
     "@sentry/nextjs": "7.46.0",
     "@trpc/server": "10.18.0",
     "@typebot.io/js": "workspace:*",
+    "@typebot.io/prisma": "workspace:*",
     "@typebot.io/react": "workspace:*",
     "aws-sdk": "2.1348.0",
     "bot-engine": "workspace:*",
     "cors": "2.8.5",
-    "@typebot.io/prisma": "workspace:*",
     "google-spreadsheet": "3.3.0",
     "got": "12.6.0",
     "libphonenumber-js": "1.10.24",
@@ -39,6 +40,10 @@
     "@faire/mjml-react": "3.2.0",
     "@paralleldrive/cuid2": "2.2.0",
     "@playwright/test": "1.32.1",
+    "@typebot.io/emails": "workspace:*",
+    "@typebot.io/lib": "workspace:*",
+    "@typebot.io/schemas": "workspace:*",
+    "@typebot.io/tsconfig": "workspace:*",
     "@types/cors": "2.8.13",
     "@types/google-spreadsheet": "3.3.1",
     "@types/node": "18.15.11",
@@ -48,17 +53,13 @@
     "@types/react": "18.0.32",
     "@types/sanitize-html": "2.9.0",
     "dotenv": "16.0.3",
-    "@typebot.io/emails": "workspace:*",
     "eslint": "8.37.0",
     "eslint-config-custom": "workspace:*",
     "google-auth-library": "8.7.0",
-    "@typebot.io/schemas": "workspace:*",
     "node-fetch": "3.3.1",
     "papaparse": "5.4.1",
     "superjson": "1.12.2",
-    "@typebot.io/tsconfig": "workspace:*",
     "typescript": "5.0.3",
-    "@typebot.io/lib": "workspace:*",
     "zod": "3.21.4"
   }
 }

apps/viewer/src/features/blocks/integrations/openai/createChatCompletionOpenAI.ts

@@ -11,6 +11,7 @@ import {
 import {
   ChatCompletionOpenAIOptions,
   OpenAICredentials,
+  modelLimit,
 } from '@typebot.io/schemas/features/blocks/integrations/openai'
 import { OpenAIApi, Configuration, ChatCompletionRequestMessage } from 'openai'
 import { isDefined, byId, isNotEmpty, isEmpty } from '@typebot.io/lib'
@@ -20,6 +21,9 @@ import { updateVariables } from '@/features/variables/updateVariables'
 import { parseVariables } from '@/features/variables/parseVariables'
 import { saveSuccessLog } from '@/features/logs/saveSuccessLog'
 import { parseVariableNumber } from '@/features/variables/parseVariableNumber'
+import { encoding_for_model } from '@dqbd/tiktoken'
+
+const minTokenCompletion = 200
 
 export const createChatCompletionOpenAI = async (
   state: SessionState,
@@ -56,7 +60,8 @@ export const createChatCompletionOpenAI = async (
     apiKey,
   })
   const { variablesTransformedToList, messages } = parseMessages(
-    newSessionState.typebot.variables
+    newSessionState.typebot.variables,
+    options.model
   )(options.messages)
   if (variablesTransformedToList.length > 0)
     newSessionState = await updateVariables(state)(variablesTransformedToList)
@@ -148,16 +153,19 @@ export const createChatCompletionOpenAI = async (
 }
 
 const parseMessages =
-  (variables: Variable[]) =>
+  (variables: Variable[], model: ChatCompletionOpenAIOptions['model']) =>
   (
     messages: ChatCompletionOpenAIOptions['messages']
   ): {
     variablesTransformedToList: VariableWithValue[]
     messages: ChatCompletionRequestMessage[]
   } => {
     const variablesTransformedToList: VariableWithValue[] = []
+    const firstMessagesSequenceIndex = messages.findIndex(
+      (message) => message.role === 'Messages sequence ✨'
+    )
     const parsedMessages = messages
-      .flatMap((message) => {
+      .flatMap((message, index) => {
         if (!message.role) return
         if (message.role === 'Messages sequence ✨') {
           if (
@@ -189,23 +197,51 @@ const parseMessages =
               variable.id === message.content?.assistantMessagesVariableId
           )?.value ?? []) as string[]
 
+          let allMessages: ChatCompletionRequestMessage[] = []
+
           if (userMessages.length > assistantMessages.length)
-            return userMessages.flatMap((userMessage, index) => [
+            allMessages = userMessages.flatMap((userMessage, index) => [
               {
                 role: 'user',
                 content: userMessage,
               },
-              { role: 'assistant', content: assistantMessages[index] },
+              { role: 'assistant', content: assistantMessages.at(index) ?? '' },
             ]) satisfies ChatCompletionRequestMessage[]
           else {
-            return assistantMessages.flatMap((assistantMessage, index) => [
-              { role: 'assistant', content: assistantMessage },
-              {
-                role: 'user',
-                content: userMessages[index],
-              },
-            ]) satisfies ChatCompletionRequestMessage[]
+            allMessages = assistantMessages.flatMap(
+              (assistantMessage, index) => [
+                { role: 'assistant', content: assistantMessage },
+                {
+                  role: 'user',
+                  content: userMessages.at(index) ?? '',
+                },
+              ]
+            ) satisfies ChatCompletionRequestMessage[]
           }
+
+          if (index !== firstMessagesSequenceIndex) return allMessages
+
+          const encoder = encoding_for_model(model)
+          let messagesToSend: ChatCompletionRequestMessage[] = []
+          let tokenCount = 0
+
+          for (let i = allMessages.length - 1; i >= 0; i--) {
+            const message = allMessages[i]
+            const tokens = encoder.encode(message.content)
+
+            if (
+              tokenCount + tokens.length - minTokenCompletion >
+              modelLimit[model]
+            ) {
+              break
+            }
+            tokenCount += tokens.length
+            messagesToSend = [message, ...messagesToSend]
+          }
+
+          encoder.free()
+
+          return messagesToSend
         }
         return {
           role: message.role,

packages/schemas/features/blocks/integrations/openai.ts

@@ -14,6 +14,15 @@ export const chatCompletionModels = [
   'gpt-3.5-turbo-0301',
 ] as const
 
+export const modelLimit = {
+  'gpt-3.5-turbo': 4096,
+  'gpt-3.5-turbo-0301': 4096,
+  'gpt-4': 8192,
+  'gpt-4-0314': 8192,
+  'gpt-4-32k': 32768,
+  'gpt-4-32k-0314': 32768,
+} as const
+
 export const chatCompletionMessageRoles = [
   'system',
   'user',