snow-ai 0.3.7 → 0.3.8

package/dist/agents/compactAgent.js

@@ -101,19 +101,22 @@ export class CompactAgent {
                         model: this.modelName,
                         messages,
                         max_tokens: 4096,
+                        includeBuiltinSystemPrompt: false, // 不需要内置系统提示词
                     }, abortSignal);
                     break;
                 case 'gemini':
                     streamGenerator = createStreamingGeminiCompletion({
                         model: this.modelName,
-                        messages
+                        messages,
+                        includeBuiltinSystemPrompt: false, // 不需要内置系统提示词
                     }, abortSignal);
                     break;
                 case 'responses':
                     streamGenerator = createStreamingResponse({
                         model: this.modelName,
                         messages,
-                        stream: true
+                        stream: true,
+                        includeBuiltinSystemPrompt: false, // 不需要内置系统提示词
                     }, abortSignal);
                     break;
                 case 'chat':
@@ -121,7 +124,8 @@ export class CompactAgent {
                     streamGenerator = createStreamingChatCompletion({
                         model: this.modelName,
                         messages,
-                        stream: true
+                        stream: true,
+                        includeBuiltinSystemPrompt: false, // 不需要内置系统提示词
                     }, abortSignal);
                     break;
             }

package/dist/agents/summaryAgent.d.ts

@@ -0,0 +1,57 @@
+/**
+ * Summary Agent Service
+ *
+ * Generates concise summaries for conversations after the first user-assistant exchange.
+ * This service operates in the background without blocking the main conversation flow.
+ *
+ * Features:
+ * - Uses basicModel for efficient, low-cost summarization
+ * - Follows the same API routing as main flow (chat, responses, gemini, anthropic)
+ * - Generates title (max 50 chars) and summary (max 150 chars)
+ * - Only runs once after the first complete conversation exchange
+ * - Silent execution with error handling to prevent main flow disruption
+ */
+export declare class SummaryAgent {
+    private modelName;
+    private requestMethod;
+    private initialized;
+    /**
+     * Initialize the summary agent with current configuration
+     * @returns true if initialized successfully, false otherwise
+     */
+    private initialize;
+    /**
+     * Check if summary agent is available
+     */
+    isAvailable(): Promise<boolean>;
+    /**
+     * Call the model with streaming API and assemble complete response
+     * Uses the same routing logic as main flow for consistency
+     *
+     * @param messages - Chat messages
+     * @param abortSignal - Optional abort signal to cancel the request
+     */
+    private callModel;
+    /**
+     * Generate title and summary for a conversation
+     *
+     * @param userMessage - User's first message content
+     * @param assistantMessage - Assistant's first response content
+     * @param abortSignal - Optional abort signal to cancel generation
+     * @returns Object containing title and summary, or null if generation fails
+     */
+    generateSummary(userMessage: string, assistantMessage: string, abortSignal?: AbortSignal): Promise<{
+        title: string;
+        summary: string;
+    } | null>;
+    /**
+     * Generate fallback summary when AI generation fails
+     * Simply truncates the user message for title and summary
+     */
+    private generateFallbackSummary;
+    /**
+     * Truncate string to specified length, adding ellipsis if truncated
+     */
+    private truncateString;
+}
+export declare const summaryAgent: SummaryAgent;

package/dist/agents/summaryAgent.js

@@ -0,0 +1,259 @@
+import { getOpenAiConfig } from '../utils/apiConfig.js';
+import { logger } from '../utils/logger.js';
+import { createStreamingChatCompletion } from '../api/chat.js';
+import { createStreamingResponse } from '../api/responses.js';
+import { createStreamingGeminiCompletion } from '../api/gemini.js';
+import { createStreamingAnthropicCompletion } from '../api/anthropic.js';
+/**
+ * Summary Agent Service
+ *
+ * Generates concise summaries for conversations after the first user-assistant exchange.
+ * This service operates in the background without blocking the main conversation flow.
+ *
+ * Features:
+ * - Uses basicModel for efficient, low-cost summarization
+ * - Follows the same API routing as main flow (chat, responses, gemini, anthropic)
+ * - Generates title (max 50 chars) and summary (max 150 chars)
+ * - Only runs once after the first complete conversation exchange
+ * - Silent execution with error handling to prevent main flow disruption
+ */
+export class SummaryAgent {
+    constructor() {
+        Object.defineProperty(this, "modelName", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: ''
+        });
+        Object.defineProperty(this, "requestMethod", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: 'chat'
+        });
+        Object.defineProperty(this, "initialized", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: false
+        });
+    }
+    /**
+     * Initialize the summary agent with current configuration
+     * @returns true if initialized successfully, false otherwise
+     */
+    async initialize() {
+        try {
+            const config = getOpenAiConfig();
+            // Check if basic model is configured
+            if (!config.basicModel) {
+                logger.warn('Summary agent: Basic model not configured, using advanced model as fallback');
+                // Fallback to advanced model if basic model is not configured
+                if (!config.advancedModel) {
+                    logger.warn('Summary agent: No model configured');
+                    return false;
+                }
+                this.modelName = config.advancedModel;
+            }
+            else {
+                this.modelName = config.basicModel;
+            }
+            this.requestMethod = config.requestMethod;
+            this.initialized = true;
+            return true;
+        }
+        catch (error) {
+            logger.warn('Summary agent: Failed to initialize:', error);
+            return false;
+        }
+    }
+    /**
+     * Check if summary agent is available
+     */
+    async isAvailable() {
+        if (!this.initialized) {
+            return await this.initialize();
+        }
+        return true;
+    }
+    /**
+     * Call the model with streaming API and assemble complete response
+     * Uses the same routing logic as main flow for consistency
+     *
+     * @param messages - Chat messages
+     * @param abortSignal - Optional abort signal to cancel the request
+     */
+    async callModel(messages, abortSignal) {
+        let streamGenerator;
+        // Route to appropriate streaming API based on request method
+        switch (this.requestMethod) {
+            case 'anthropic':
+                streamGenerator = createStreamingAnthropicCompletion({
+                    model: this.modelName,
+                    messages,
+                    max_tokens: 500, // Limited tokens for summary generation
+                    includeBuiltinSystemPrompt: false, // 不需要内置系统提示词
+                }, abortSignal);
+                break;
+            case 'gemini':
+                streamGenerator = createStreamingGeminiCompletion({
+                    model: this.modelName,
+                    messages,
+                    includeBuiltinSystemPrompt: false, // 不需要内置系统提示词
+                }, abortSignal);
+                break;
+            case 'responses':
+                streamGenerator = createStreamingResponse({
+                    model: this.modelName,
+                    messages,
+                    stream: true,
+                    includeBuiltinSystemPrompt: false, // 不需要内置系统提示词
+                }, abortSignal);
+                break;
+            case 'chat':
+            default:
+                streamGenerator = createStreamingChatCompletion({
+                    model: this.modelName,
+                    messages,
+                    stream: true,
+                    includeBuiltinSystemPrompt: false, // 不需要内置系统提示词
+                }, abortSignal);
+                break;
+        }
+        // Assemble complete content from streaming response
+        let completeContent = '';
+        try {
+            for await (const chunk of streamGenerator) {
+                // Check abort signal
+                if (abortSignal?.aborted) {
+                    throw new Error('Request aborted');
+                }
+                // Handle different chunk formats based on request method
+                if (this.requestMethod === 'chat') {
+                    // Chat API uses standard OpenAI format
+                    if (chunk.choices && chunk.choices[0]?.delta?.content) {
+                        completeContent += chunk.choices[0].delta.content;
+                    }
+                }
+                else {
+                    // Responses, Gemini, and Anthropic APIs use unified format
+                    if (chunk.type === 'content' && chunk.content) {
+                        completeContent += chunk.content;
+                    }
+                }
+            }
+        }
+        catch (streamError) {
+            logger.error('Summary agent: Streaming error:', streamError);
+            throw streamError;
+        }
+        return completeContent;
+    }
+    /**
+     * Generate title and summary for a conversation
+     *
+     * @param userMessage - User's first message content
+     * @param assistantMessage - Assistant's first response content
+     * @param abortSignal - Optional abort signal to cancel generation
+     * @returns Object containing title and summary, or null if generation fails
+     */
+    async generateSummary(userMessage, assistantMessage, abortSignal) {
+        const available = await this.isAvailable();
+        if (!available) {
+            logger.warn('Summary agent: Not available, using fallback summary');
+            return this.generateFallbackSummary(userMessage, assistantMessage);
+        }
+        try {
+            const summaryPrompt = `You are a conversation summarization assistant. Based on the first exchange between the user and AI assistant below, generate a concise title and summary.
+
+IMPORTANT: Generate the title and summary in the SAME LANGUAGE as the user's message. If the user writes in Chinese, respond in Chinese. If in English, respond in English.
+
+User message:
+${userMessage}
+
+AI assistant reply:
+${assistantMessage}
+
+Requirements:
+1. Generate a short title (max 50 characters) that captures the conversation topic
+2. Generate a summary (max 150 characters) that briefly describes the core content
+3. Title should be concise and clear, avoid complete sentences
+4. Summary should contain key information while staying brief
+5. Use the SAME LANGUAGE as the user's message
+
+Output in the following JSON format (JSON only, no other content):
+{
+  "title": "Conversation title",
+  "summary": "Conversation summary"
+}`;
+            const messages = [
+                {
+                    role: 'user',
+                    content: summaryPrompt,
+                },
+            ];
+            const response = await this.callModel(messages, abortSignal);
+            if (!response || response.trim().length === 0) {
+                logger.warn('Summary agent: Empty response, using fallback');
+                return this.generateFallbackSummary(userMessage, assistantMessage);
+            }
+            // Parse JSON response
+            try {
+                // Extract JSON from markdown code blocks if present
+                let jsonStr = response.trim();
+                const jsonMatch = jsonStr.match(/```(?:json)?\s*\n?([\s\S]*?)\n?```/);
+                if (jsonMatch) {
+                    jsonStr = jsonMatch[1].trim();
+                }
+                const parsed = JSON.parse(jsonStr);
+                if (!parsed.title || !parsed.summary) {
+                    logger.warn('Summary agent: Invalid JSON structure, using fallback');
+                    return this.generateFallbackSummary(userMessage, assistantMessage);
+                }
+                // Ensure title and summary are within length limits
+                const title = this.truncateString(parsed.title, 50);
+                const summary = this.truncateString(parsed.summary, 150);
+                logger.info('Summary agent: Successfully generated summary', {
+                    title,
+                    summary,
+                });
+                return { title, summary };
+            }
+            catch (parseError) {
+                logger.warn('Summary agent: Failed to parse JSON response, using fallback', parseError);
+                return this.generateFallbackSummary(userMessage, assistantMessage);
+            }
+        }
+        catch (error) {
+            logger.error('Summary agent: Failed to generate summary', error);
+            return this.generateFallbackSummary(userMessage, assistantMessage);
+        }
+    }
+    /**
+     * Generate fallback summary when AI generation fails
+     * Simply truncates the user message for title and summary
+     */
+    generateFallbackSummary(userMessage, _assistantMessage) {
+        // Clean newlines and extra spaces
+        const cleanedUser = userMessage
+            .replace(/[\r\n]+/g, ' ')
+            .replace(/\s+/g, ' ')
+            .trim();
+        // Use first 50 chars as title
+        const title = this.truncateString(cleanedUser, 50);
+        // Use first 150 chars as summary
+        const summary = this.truncateString(cleanedUser, 150);
+        return { title, summary };
+    }
+    /**
+     * Truncate string to specified length, adding ellipsis if truncated
+     */
+    truncateString(str, maxLength) {
+        if (str.length <= maxLength) {
+            return str;
+        }
+        return str.slice(0, maxLength - 3) + '...';
+    }
+}
+// Export singleton instance
+export const summaryAgent = new SummaryAgent();

package/dist/api/anthropic.d.ts

@@ -6,6 +6,7 @@ export interface AnthropicOptions {
     max_tokens?: number;
     tools?: ChatCompletionTool[];
     sessionId?: string;
+    includeBuiltinSystemPrompt?: boolean;
 }
 export interface AnthropicStreamChunk {
     type: 'content' | 'tool_calls' | 'tool_call_delta' | 'done' | 'usage';

package/dist/api/anthropic.js

@@ -68,8 +68,10 @@ function convertToolsToAnthropic(tools) {
 /**
  * Convert our ChatMessage format to Anthropic's message format
  * Adds cache_control to system prompt and last user message for prompt caching
+ * @param messages - The messages to convert
+ * @param includeBuiltinSystemPrompt - Whether to include builtin system prompt (default true)
  */
-function convertToAnthropicMessages(messages) {
+function convertToAnthropicMessages(messages, includeBuiltinSystemPrompt = true) {
     const customSystemPrompt = getCustomSystemPrompt();
     let systemContent;
     const anthropicMessages = [];
@@ -149,20 +151,25 @@ function convertToAnthropicMessages(messages) {
             });
         }
     }
+    // 如果配置了自定义系统提示词（最高优先级，始终添加）
     if (customSystemPrompt) {
         systemContent = customSystemPrompt;
-        anthropicMessages.unshift({
-            role: 'user',
-            content: [
-                {
-                    type: 'text',
-                    text: getSystemPrompt(),
-                    cache_control: { type: 'ephemeral' },
-                },
-            ],
-        });
+        if (includeBuiltinSystemPrompt) {
+            // 将默认系统提示词作为第一条用户消息
+            anthropicMessages.unshift({
+                role: 'user',
+                content: [
+                    {
+                        type: 'text',
+                        text: getSystemPrompt(),
+                        cache_control: { type: 'ephemeral' },
+                    },
+                ],
+            });
+        }
     }
-    else if (!systemContent) {
+    else if (!systemContent && includeBuiltinSystemPrompt) {
+        // 没有自定义系统提示词，但需要添加默认系统提示词
         systemContent = getSystemPrompt();
     }
     let lastUserMessageIndex = -1;
@@ -253,7 +260,7 @@ async function* parseSSEStream(reader) {
 export async function* createStreamingAnthropicCompletion(options, abortSignal, onRetry) {
     yield* withRetryGenerator(async function* () {
         const config = getAnthropicConfig();
-        const { system, messages } = convertToAnthropicMessages(options.messages);
+        const { system, messages } = convertToAnthropicMessages(options.messages, options.includeBuiltinSystemPrompt !== false);
         const sessionId = options.sessionId || randomUUID();
         const userId = generateUserId(sessionId);
         const requestBody = {

package/dist/api/chat.d.ts

@@ -13,6 +13,7 @@ export interface ChatCompletionOptions {
             name: string;
         };
     };
+    includeBuiltinSystemPrompt?: boolean;
 }
 export interface ChatCompletionChunk {
     id: string;

package/dist/api/chat.js

@@ -9,8 +9,10 @@ import { saveUsageToFile } from '../utils/usageLogger.js';
  * Logic:
  * 1. If custom system prompt exists: use custom as system, prepend default as first user message
  * 2. If no custom system prompt: use default as system
+ * @param messages - The messages to convert
+ * @param includeBuiltinSystemPrompt - Whether to include builtin system prompt (default true)
  */
-function convertToOpenAIMessages(messages, includeSystemPrompt = true) {
+function convertToOpenAIMessages(messages, includeBuiltinSystemPrompt = true) {
     const customSystemPrompt = getCustomSystemPrompt();
     let result = messages.map(msg => {
         // 如果消息包含图片，使用 content 数组格式
@@ -56,14 +58,13 @@ function convertToOpenAIMessages(messages, includeSystemPrompt = true) {
         }
         return baseMessage;
     });
-    // 如果需要系统提示词
-    if (includeSystemPrompt) {
-        // 如果第一条消息已经是 system 消息，跳过
-        if (result.length > 0 && result[0]?.role === 'system') {
-            return result;
-        }
-        // 如果配置了自定义系统提示词
-        if (customSystemPrompt) {
+    // 如果第一条消息已经是 system 消息，跳过
+    if (result.length > 0 && result[0]?.role === 'system') {
+        return result;
+    }
+    // 如果配置了自定义系统提示词（最高优先级，始终添加）
+    if (customSystemPrompt) {
+        if (includeBuiltinSystemPrompt) {
             // 自定义系统提示词作为 system 消息，默认系统提示词作为第一条 user 消息
             result = [
                 {
@@ -78,16 +79,26 @@ function convertToOpenAIMessages(messages, includeSystemPrompt = true) {
             ];
         }
         else {
-            // 没有自定义系统提示词，默认系统提示词作为 system 消息
+            // 只添加自定义系统提示词
             result = [
                 {
                     role: 'system',
-                    content: getSystemPrompt(),
+                    content: customSystemPrompt,
                 },
                 ...result,
             ];
         }
     }
+    else if (includeBuiltinSystemPrompt) {
+        // 没有自定义系统提示词，但需要添加默认系统提示词
+        result = [
+            {
+                role: 'system',
+                content: getSystemPrompt(),
+            },
+            ...result,
+        ];
+    }
     return result;
 }
 let openaiConfig = null;
@@ -161,7 +172,7 @@ export async function* createStreamingChatCompletion(options, abortSignal, onRet
     yield* withRetryGenerator(async function* () {
         const requestBody = {
             model: options.model,
-            messages: convertToOpenAIMessages(options.messages),
+            messages: convertToOpenAIMessages(options.messages, options.includeBuiltinSystemPrompt !== false),
             stream: true,
             stream_options: { include_usage: true },
             temperature: options.temperature || 0.7,

package/dist/api/gemini.d.ts

@@ -4,6 +4,7 @@ export interface GeminiOptions {
     messages: ChatMessage[];
     temperature?: number;
     tools?: ChatCompletionTool[];
+    includeBuiltinSystemPrompt?: boolean;
 }
 export interface GeminiStreamChunk {
     type: 'content' | 'tool_calls' | 'tool_call_delta' | 'done' | 'usage';

package/dist/api/gemini.js

@@ -53,8 +53,10 @@ function convertToolsToGemini(tools) {
 }
 /**
  * Convert our ChatMessage format to Gemini's format
+ * @param messages - The messages to convert
+ * @param includeBuiltinSystemPrompt - Whether to include builtin system prompt (default true)
  */
-function convertToGeminiMessages(messages) {
+function convertToGeminiMessages(messages, includeBuiltinSystemPrompt = true) {
     const customSystemPrompt = getCustomSystemPrompt();
     let systemInstruction;
     const contents = [];
@@ -197,15 +199,19 @@ function convertToGeminiMessages(messages) {
         contents.push({ role, parts });
     }
     // Handle system instruction
+    // 如果配置了自定义系统提示词（最高优先级，始终添加）
     if (customSystemPrompt) {
         systemInstruction = customSystemPrompt;
-        // Prepend default system prompt as first user message
-        contents.unshift({
-            role: 'user',
-            parts: [{ text: getSystemPrompt() }],
-        });
+        if (includeBuiltinSystemPrompt) {
+            // Prepend default system prompt as first user message
+            contents.unshift({
+                role: 'user',
+                parts: [{ text: getSystemPrompt() }],
+            });
+        }
     }
-    else if (!systemInstruction) {
+    else if (!systemInstruction && includeBuiltinSystemPrompt) {
+        // 没有自定义系统提示词，但需要添加默认系统提示词
         systemInstruction = getSystemPrompt();
     }
     return { systemInstruction, contents };
@@ -217,7 +223,7 @@ export async function* createStreamingGeminiCompletion(options, abortSignal, onR
     const config = getGeminiConfig();
     // 使用重试包装生成器
     yield* withRetryGenerator(async function* () {
-        const { systemInstruction, contents } = convertToGeminiMessages(options.messages);
+        const { systemInstruction, contents } = convertToGeminiMessages(options.messages, options.includeBuiltinSystemPrompt !== false);
         // Build request payload
         const requestBody = {
             contents,

package/dist/api/responses.d.ts

@@ -14,6 +14,7 @@ export interface ResponseOptions {
     prompt_cache_key?: string;
     store?: boolean;
     include?: string[];
+    includeBuiltinSystemPrompt?: boolean;
 }
 export interface ResponseStreamChunk {
     type: 'content' | 'tool_calls' | 'tool_call_delta' | 'reasoning_delta' | 'reasoning_started' | 'reasoning_data' | 'done' | 'usage';

package/dist/api/responses.js

@@ -82,7 +82,7 @@ function getOpenAIConfig() {
 export function resetOpenAIClient() {
     openaiConfig = null;
 }
-function convertToResponseInput(messages) {
+function convertToResponseInput(messages, includeBuiltinSystemPrompt = true) {
     const customSystemPrompt = getCustomSystemPrompt();
     const result = [];
     for (const msg of messages) {
@@ -160,24 +160,32 @@ function convertToResponseInput(messages) {
     }
     // 确定系统提示词：参考 anthropic.ts 的逻辑
     let systemInstructions;
+    // 如果配置了自定义系统提示词（最高优先级，始终添加）
     if (customSystemPrompt) {
-        // 有自定义系统提示词：自定义作为 instructions，默认作为第一条用户消息
+        // 有自定义系统提示词：自定义作为 instructions
         systemInstructions = customSystemPrompt;
-        result.unshift({
-            type: 'message',
-            role: 'user',
-            content: [
-                {
-                    type: 'input_text',
-                    text: getSystemPrompt(),
-                },
-            ],
-        });
+        if (includeBuiltinSystemPrompt) {
+            // 默认系统提示词作为第一条用户消息
+            result.unshift({
+                type: 'message',
+                role: 'user',
+                content: [
+                    {
+                        type: 'input_text',
+                        text: getSystemPrompt(),
+                    },
+                ],
+            });
+        }
     }
-    else {
-        // 没有自定义系统提示词：默认作为 instructions
+    else if (includeBuiltinSystemPrompt) {
+        // 没有自定义系统提示词，但需要添加默认系统提示词
         systemInstructions = getSystemPrompt();
     }
+    else {
+        // 既没有自定义系统提示词，也不需要添加默认系统提示词
+        systemInstructions = 'You are a helpful assistant.';
+    }
     return { input: result, systemInstructions };
 }
 /**
@@ -228,7 +236,7 @@ async function* parseSSEStream(reader) {
 export async function* createStreamingResponse(options, abortSignal, onRetry) {
     const config = getOpenAIConfig();
     // 提取系统提示词和转换后的消息
-    const { input: requestInput, systemInstructions } = convertToResponseInput(options.messages);
+    const { input: requestInput, systemInstructions } = convertToResponseInput(options.messages, options.includeBuiltinSystemPrompt !== false);
     // 使用重试包装生成器
     yield* withRetryGenerator(async function* () {
         const requestPayload = {