fix(server): token calculate (#12667)

packages/backend/server/src/plugins/copilot/session.ts

@@ -19,7 +19,6 @@ import { ChatMessageCache } from './message';
 import { PromptService } from './prompt';
 import { PromptMessage, PromptParams } from './providers';
 import {
-  AvailableModel,
   ChatHistory,
   ChatMessage,
   ChatMessageSchema,
@@ -38,7 +37,7 @@ export class ChatSession implements AsyncDisposable {
     private readonly messageCache: ChatMessageCache,
     private readonly state: ChatSessionState,
     private readonly dispose?: (state: ChatSessionState) => Promise<void>,
-    private readonly maxTokenSize = 3840
+    private readonly maxTokenSize = state.prompt.config?.maxTokens || 128 * 1024
   ) {}
 
   get model() {
@@ -297,8 +296,8 @@ export class ChatSessionService {
               messageCost: { increment: userMessages.length },
               tokenCost: {
                 increment: this.calculateTokenSize(
-                  userMessages,
-                  state.prompt.model as AvailableModel
+                  state.messages,
+                  state.prompt.model
                 ),
               },
             },
@@ -402,10 +401,7 @@ export class ChatSessionService {
     });
   }
 
-  private calculateTokenSize(
-    messages: PromptMessage[],
-    model: AvailableModel
-  ): number {
+  private calculateTokenSize(messages: PromptMessage[], model: string): number {
     const encoder = getTokenEncoder(model);
     return messages
       .map(m => encoder?.count(m.content) ?? 0)

packages/backend/server/src/plugins/copilot/types.ts

@@ -47,26 +47,24 @@ export enum AvailableModels {
   Gpt41 = 'gpt-4.1',
   Gpt410414 = 'gpt-4.1-2025-04-14',
   Gpt41Mini = 'gpt-4.1-mini',
+  Gpt41Nano = 'gpt-4.1-nano',
   // embeddings
   TextEmbedding3Large = 'text-embedding-3-large',
   TextEmbedding3Small = 'text-embedding-3-small',
   TextEmbeddingAda002 = 'text-embedding-ada-002',
-  // moderation
-  TextModerationLatest = 'text-moderation-latest',
-  TextModerationStable = 'text-moderation-stable',
   // text to image
   DallE3 = 'dall-e-3',
+  GptImage = 'gpt-image-1',
 }
 
-export type AvailableModel = keyof typeof AvailableModels;
+const availableModels = Object.values(AvailableModels);
 
 export function getTokenEncoder(model?: string | null): Tokenizer | null {
   if (!model) return null;
-  const modelStr = AvailableModels[model as AvailableModel];
-  if (!modelStr) return null;
-  if (modelStr.startsWith('gpt')) {
-    return fromModelName(modelStr);
-  } else if (modelStr.startsWith('dall')) {
+  if (!availableModels.includes(model as AvailableModels)) return null;
+  if (model.startsWith('gpt')) {
+    return fromModelName(model);
+  } else if (model.startsWith('dall')) {
     // dalle don't need to calc the token
     return null;
   } else {

packages/backend/server/src/schema.gql

@@ -262,12 +262,12 @@ enum CopilotModels {
   Gpt4OmniMini0718
   Gpt41
   Gpt41Mini
+  Gpt41Nano
   Gpt410414
+  GptImage
   TextEmbedding3Large
   TextEmbedding3Small
   TextEmbeddingAda002
-  TextModerationLatest
-  TextModerationStable
 }
 
 input CopilotPromptConfigInput {