smart_prompt 0.4.4 → 0.5.0

data/workers/history_management_examples.rb

@@ -0,0 +1,407 @@
+# History Management Example Workers
+# These examples demonstrate the new intelligent history management features
+# introduced in the history-optimization feature.
+#
+# Key Features Demonstrated:
+# 1. Explicit session management with unique session IDs
+# 2. Different context strategies (sliding window, relevance-based, summary-based, hybrid)
+# 3. Session operations (clear, export, search, stats, delete)
+# 4. Backward compatibility with existing API
+#
+# Requirements Validated: 5.5
+
+# ============================================================================
+# Example 1: Explicit Session Management
+# ============================================================================
+# Demonstrates: Requirement 1 (Session Isolation), Requirement 5.2 (Default Sessions)
+
+# Simple chat worker with automatic session management
+# When with_history: true is used without a session_id, a default session is created
+SmartPrompt.define_worker :basic_chat do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  sys_msg("You are a helpful AI assistant.", params)
+  prompt(params[:message], with_history: true)
+  send_msg
+end
+
+# Chat worker with explicit session management
+# Use session_id to maintain separate conversations
+# Each session maintains isolated history (Requirement 1.1-1.4)
+SmartPrompt.define_worker :explicit_session_chat do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  # Session ID can be user ID, conversation ID, or any unique identifier
+  session_id = params[:session_id] || "default_session"
+  
+  sys_msg("You are a helpful AI assistant.", params)
+  prompt(params[:message], with_history: true, session_id: session_id)
+  send_msg
+end
+
+# Multi-user chat with isolated sessions per user
+# Demonstrates: Session isolation for concurrent users
+SmartPrompt.define_worker :multi_user_chat do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  # Use user ID as session ID for isolation
+  user_id = params[:user_id] || raise("user_id is required")
+  session_id = "user_#{user_id}"
+  
+  # Configure per-user session with limits (Requirement 2.1-2.2)
+  session_config = {
+    max_messages: 50,
+    max_tokens: 3000,
+    context_strategy: :sliding_window,
+    preserve_system_messages: true
+  }
+  
+  user_name = params[:user_name] || "User"
+  sys_msg("You are a personal AI assistant for #{user_name}.", params)
+  prompt(params[:message], with_history: true, session_id: session_id)
+  
+  # Apply session configuration
+  @engine.history_manager.get_session(session_id, session_config)
+  
+  send_msg
+end
+
+# ============================================================================
+# Example 2: Different Context Strategies
+# ============================================================================
+# Demonstrates: Requirement 6.1 (Context Strategy Configuration), 
+#               Requirement 8.2 (Multiple Strategy Support)
+
+# 2.1 Sliding Window Strategy
+# Keeps most recent N messages - Best for: Real-time chat, customer support
+SmartPrompt.define_worker :sliding_window_chat do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  session_id = params[:session_id] || "sliding_window_session"
+  
+  # Configure session with sliding window strategy (Requirement 8.2)
+  session_config = {
+    max_messages: 20,
+    max_tokens: 2000,
+    context_strategy: :sliding_window,
+    preserve_system_messages: true  # Requirement 2.5
+  }
+  
+  sys_msg("You are a helpful customer support assistant.", params)
+  prompt(params[:message], with_history: true, session_id: session_id)
+  
+  # Apply session configuration
+  @engine.history_manager.get_session(session_id, session_config)
+  
+  send_msg
+end
+
+# 2.2 Relevance-Based Strategy
+# Selects semantically relevant messages - Best for: Q&A systems, knowledge bases
+SmartPrompt.define_worker :relevance_chat do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  session_id = params[:session_id] || "relevance_session"
+  
+  # Configure session with relevance-based strategy (Requirement 8.1, 8.3)
+  session_config = {
+    max_messages: 100,
+    max_tokens: 4000,
+    context_strategy: :relevance_based,
+    preserve_system_messages: true
+  }
+  
+  sys_msg("You are a knowledgeable AI assistant that provides accurate answers.", params)
+  prompt(params[:message], with_history: true, session_id: session_id)
+  
+  @engine.history_manager.get_session(session_id, session_config)
+  
+  send_msg
+end
+
+# 2.3 Summary-Based Strategy
+# Automatically compresses old messages - Best for: Long conversations, extended dialogues
+SmartPrompt.define_worker :summary_chat do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  session_id = params[:session_id] || "summary_session"
+  
+  # Configure session with summary-based strategy (Requirement 9.1-9.3)
+  session_config = {
+    max_messages: 200,
+    max_tokens: 8000,
+    context_strategy: :summary_based,
+    preserve_system_messages: true
+  }
+  
+  sys_msg("You are a thoughtful AI assistant for extended conversations.", params)
+  prompt(params[:message], with_history: true, session_id: session_id)
+  
+  @engine.history_manager.get_session(session_id, session_config)
+  
+  send_msg
+end
+
+# 2.4 Hybrid Strategy
+# Adaptively combines multiple strategies - Best for: General-purpose applications
+SmartPrompt.define_worker :hybrid_chat do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  session_id = params[:session_id] || "hybrid_session"
+  
+  # Configure session with hybrid strategy (Requirement 6.1)
+  session_config = {
+    max_messages: 150,
+    max_tokens: 6000,
+    context_strategy: :hybrid,
+    preserve_system_messages: true
+  }
+  
+  sys_msg("You are an intelligent AI assistant that adapts to conversation context.", params)
+  prompt(params[:message], with_history: true, session_id: session_id)
+  
+  @engine.history_manager.get_session(session_id, session_config)
+  
+  send_msg
+end
+
+# ============================================================================
+# Example 3: Session Operations (clear, export, search, stats, delete)
+# ============================================================================
+# Demonstrates: Requirement 4 (Session Management Operations)
+
+# Worker demonstrating all session management operations
+SmartPrompt.define_worker :session_operations do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  session_id = params[:session_id] || "default_session"
+  operation = params[:operation] # :clear, :export, :search, :stats, :delete
+  
+  case operation
+  when :clear
+    # Clear session history (keeps system messages by default) - Requirement 4.1
+    @engine.history_manager.clear_session(session_id, keep_system_messages: true)
+    "Session cleared successfully"
+    
+  when :export
+    # Export session data in structured format - Requirement 4.2
+    @engine.history_manager.export_session(session_id, format: :json)
+    
+  when :search
+    # Search for messages containing specific text - Requirement 4.3
+    query = params[:query]
+    results = @engine.history_manager.search_messages(session_id, query)
+    "Found #{results.count} messages matching '#{query}'"
+    
+  when :stats
+    # Get session statistics (message count, token usage) - Requirement 4.4
+    @engine.history_manager.get_stats(session_id)
+    
+  when :delete
+    # Delete session completely (memory and disk) - Requirement 4.5
+    @engine.history_manager.delete_session(session_id)
+    "Session deleted successfully"
+    
+  else
+    "Unknown operation: #{operation}"
+  end
+end
+
+# Example: Get system-wide statistics
+SmartPrompt.define_worker :system_stats do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  # Get system-wide statistics - Requirement 11.3
+  stats = @engine.history_manager.get_stats
+  
+  {
+    active_sessions: stats[:active_sessions],
+    total_messages: stats[:total_messages],
+    total_tokens: stats[:total_tokens],
+    cache_hit_rate: stats[:cache_hit_rate],
+    messages_added: stats[:messages_added],
+    context_retrievals: stats[:context_retrievals]
+  }
+end
+
+# Example: Export metrics in Prometheus format
+SmartPrompt.define_worker :export_metrics do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  # Export metrics in Prometheus format - Requirement 11.5
+  @engine.history_manager.export_metrics(format: :prometheus)
+end
+
+# ============================================================================
+# Example 4: Backward Compatibility
+# ============================================================================
+# Demonstrates: Requirement 5 (Backward Compatibility)
+
+# Old-style worker definition (still works!) - Requirement 5.1, 5.4
+SmartPrompt.define_worker :legacy_chat do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  sys_msg("You are a helpful AI assistant.", params)
+  
+  # Old-style history usage - still works!
+  # Automatically creates a default session - Requirement 5.2
+  prompt(params[:message], with_history: true)
+  
+  send_msg(with_history: true)
+end
+
+# Worker showing compatibility mode - Requirement 5.3
+SmartPrompt.define_worker :compatibility_chat do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  sys_msg("You are a helpful AI assistant.", params)
+  
+  # Works with both old and new API
+  if params[:session_id]
+    # New API: explicit session management
+    prompt(params[:message], with_history: true, session_id: params[:session_id])
+  else
+    # Old API: automatic default session
+    prompt(params[:message], with_history: true)
+  end
+  
+  send_msg
+end
+
+# ============================================================================
+# Example 5: Advanced Use Cases
+# ============================================================================
+
+# Streaming with history management
+SmartPrompt.define_worker :streaming_history_chat do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  session_id = params[:session_id] || "streaming_session"
+  
+  sys_msg("You are a helpful AI assistant.", params)
+  prompt(params[:message], with_history: true, session_id: session_id)
+  
+  # Use send_msg_by_stream for streaming responses
+  # History is still maintained automatically
+  send_msg_by_stream(params)
+end
+
+# Custom configuration for specific use cases
+SmartPrompt.define_worker :custom_config_chat do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  session_id = params[:session_id] || "custom_session"
+  
+  # Fine-tuned configuration for a code review assistant
+  session_config = {
+    max_messages: 100,
+    max_tokens: 8000,
+    context_strategy: :relevance_based,
+    preserve_system_messages: true
+  }
+  
+  sys_msg("You are an expert code reviewer. Analyze code and provide constructive feedback.", params)
+  prompt(params[:message], with_history: true, session_id: session_id)
+  
+  @engine.history_manager.get_session(session_id, session_config)
+  
+  send_msg
+end
+
+# Persistence example - sessions are automatically saved to disk (Requirement 3.1-3.3)
+SmartPrompt.define_worker :persistent_chat do
+  use "deepseek_anthropic"
+  model "deepseek-chat"
+  
+  session_id = params[:session_id] || "persistent_session"
+  
+  # Sessions are automatically persisted to disk
+  # They will be restored on next access
+  sys_msg("You are a helpful AI assistant. Your conversations are saved.", params)
+  prompt(params[:message], with_history: true, session_id: session_id)
+  
+  send_msg
+end
+
+# ============================================================================
+# Usage Examples
+# ============================================================================
+# 
+# Example 1: Basic usage with automatic session
+#   SmartPrompt.run_worker(:basic_chat, { message: "Hello!" })
+#
+# Example 2: Explicit session management
+#   SmartPrompt.run_worker(:explicit_session_chat, { 
+#     session_id: "user_123", 
+#     message: "Hello!" 
+#   })
+#
+# Example 3: Multi-user chat
+#   SmartPrompt.run_worker(:multi_user_chat, { 
+#     user_id: "001", 
+#     user_name: "Alice",
+#     message: "Hi!" 
+#   })
+#
+# Example 4: Using different strategies
+#   SmartPrompt.run_worker(:sliding_window_chat, { message: "Hello!" })
+#   SmartPrompt.run_worker(:relevance_chat, { message: "Hello!" })
+#   SmartPrompt.run_worker(:summary_chat, { message: "Hello!" })
+#   SmartPrompt.run_worker(:hybrid_chat, { message: "Hello!" })
+#
+# Example 5: Session operations
+#   # Get statistics
+#   SmartPrompt.run_worker(:session_operations, { 
+#     session_id: "user_123", 
+#     operation: :stats 
+#   })
+#   
+#   # Search messages
+#   SmartPrompt.run_worker(:session_operations, { 
+#     session_id: "user_123", 
+#     operation: :search,
+#     query: "Ruby" 
+#   })
+#   
+#   # Export session
+#   SmartPrompt.run_worker(:session_operations, { 
+#     session_id: "user_123", 
+#     operation: :export 
+#   })
+#   
+#   # Clear session
+#   SmartPrompt.run_worker(:session_operations, { 
+#     session_id: "user_123", 
+#     operation: :clear 
+#   })
+#   
+#   # Delete session
+#   SmartPrompt.run_worker(:session_operations, { 
+#     session_id: "user_123", 
+#     operation: :delete 
+#   })
+#
+# Example 6: System-wide statistics
+#   SmartPrompt.run_worker(:system_stats, {})
+#
+# Example 7: Export Prometheus metrics
+#   SmartPrompt.run_worker(:export_metrics, {})
+#
+# Example 8: Backward compatibility (old API)
+#   SmartPrompt.run_worker(:legacy_chat, { message: "Hello!" })
+#
+# ============================================================================

data/workers/image_generation_workers.rb

@@ -0,0 +1,119 @@
+# Image Generation Workers for SmartPrompt
+#
+# These workers wrap the SiliconFlow image generation API exposed by
+# ImageGenerationAdapter (POST /v1/images/generations). All of them expect an
+# `image_gen` LLM to be configured in the engine, e.g.:
+#
+#   image_gen:
+#     adapter: "image_generation"
+#     url: "https://api.siliconflow.cn/v1/"
+#     api_key: ENV["SILICONFLOW_API_KEY"]
+#     model: "Kwai-Kolors/Kolors"
+
+# Text-to-image generation worker.
+#
+# Recognized params:
+#   prompt:, model:, negative_prompt:,
+#   image_size: (alias size:), batch_size: (alias n:),
+#   seed:, num_inference_steps:, guidance_scale:, cfg:,
+#   save_to_file:, output_dir:, filename_prefix:
+SmartPrompt.define_worker :image_generator do
+  use "image_gen"
+
+  images = generate_image(params[:prompt], {
+    model: params[:model],
+    negative_prompt: params[:negative_prompt],
+    image_size: params[:image_size] || params[:size],
+    batch_size: params[:batch_size] || params[:n],
+    seed: params[:seed],
+    num_inference_steps: params[:num_inference_steps],
+    guidance_scale: params[:guidance_scale],
+    cfg: params[:cfg],
+  })
+
+  if params[:save_to_file]
+    saved_files = save_image(images, params[:output_dir] || "./generated_images", params[:filename_prefix] || "generated")
+    { images: images, saved_files: saved_files }
+  else
+    images
+  end
+end
+
+# Image editing / image-to-image worker (for Qwen/Qwen-Image-Edit-* and similar).
+#
+# Recognized params:
+#   prompt:, model:, image: (alias image_file:), image2:, image3:,
+#   negative_prompt:, seed:, cfg:,
+#   save_to_file:, output_dir:, filename_prefix:
+SmartPrompt.define_worker :image_editor do
+  use "image_gen"
+
+  images = edit_image(params[:prompt], {
+    model: params[:model],
+    image: params[:image] || params[:image_file],
+    image2: params[:image2],
+    image3: params[:image3],
+    negative_prompt: params[:negative_prompt],
+    seed: params[:seed],
+    cfg: params[:cfg],
+  })
+
+  if params[:save_to_file]
+    saved_files = save_image(images, params[:output_dir] || "./edited_images", params[:filename_prefix] || "edited")
+    { images: images, saved_files: saved_files }
+  else
+    images
+  end
+end
+
+# Creative art generation worker. Appends the requested art style to the prompt
+# before generating.
+SmartPrompt.define_worker :art_generator do
+  use "image_gen"
+
+  prompt = params[:prompt] || "A beautiful artistic creation"
+  prompt = "#{prompt}, in the style of #{params[:art_style]}" if params[:art_style]
+
+  images = generate_image(prompt, {
+    model: params[:model],
+    negative_prompt: params[:negative_prompt],
+    image_size: params[:image_size] || params[:size],
+    batch_size: params[:batch_size] || params[:n],
+    seed: params[:seed],
+    num_inference_steps: params[:num_inference_steps],
+    guidance_scale: params[:guidance_scale],
+    cfg: params[:cfg],
+  })
+
+  if params[:save_to_file]
+    saved_files = save_image(images, params[:output_dir] || "./art_images", params[:filename_prefix] || "art")
+    { images: images, saved_files: saved_files }
+  else
+    images
+  end
+end
+
+# Product photography worker. Enhances the prompt with studio-photo wording.
+SmartPrompt.define_worker :product_image_generator do
+  use "image_gen"
+
+  prompt = "Professional product photography, studio lighting, clean background, high detail, #{params[:prompt]}"
+
+  images = generate_image(prompt, {
+    model: params[:model],
+    negative_prompt: params[:negative_prompt] || "blurry, low quality, distorted",
+    image_size: params[:image_size] || params[:size],
+    batch_size: params[:batch_size] || params[:n],
+    seed: params[:seed],
+    num_inference_steps: params[:num_inference_steps],
+    guidance_scale: params[:guidance_scale],
+    cfg: params[:cfg],
+  })
+
+  if params[:save_to_file]
+    saved_files = save_image(images, params[:output_dir] || "./product_images", params[:filename_prefix] || "product")
+    { images: images, saved_files: saved_files }
+  else
+    images
+  end
+end

data/workers/multimodal_workers.rb

@@ -0,0 +1,110 @@
+# Multimodal Workers for SmartPrompt
+# These workers demonstrate the new multimodal capabilities
+
+# Image analysis worker
+SmartPrompt.define_worker :image_analyzer do
+  use "qwen_vl"
+  model "Qwen/Qwen2.5-VL-7B-Instruct"
+
+  messages = [
+    {
+      role: "user",
+      content: [
+        { type: "text", text: params[:question] },
+        { type: "image_url", image_url: { url: params[:image_url], detail: params[:detail] || "auto" } }
+      ]
+    }
+  ]
+
+  sys_msg("你是一个专业的图像分析助手，能够准确描述和分析图像内容。", params)
+  params.merge(messages: messages)
+  send_msg
+end
+
+# Video analysis worker
+SmartPrompt.define_worker :video_analyzer do
+  use "qwen_vl"
+  model "Qwen/Qwen2.5-VL-7B-Instruct"
+
+  messages = [
+    {
+      role: "user",
+      content: [
+        { type: "text", text: params[:question] },
+        {
+          type: "video_url",
+          video_url: {
+            url: params[:video_url],
+            detail: params[:detail] || "auto",
+            max_frames: params[:max_frames] || 10,
+            fps: params[:fps] || 1
+          }
+        }
+      ]
+    }
+  ]
+
+  sys_msg("你是一个专业的视频分析助手，能够准确描述和分析视频内容。", params)
+  params.merge(messages: messages)
+  send_msg
+end
+
+# Multiple images comparison worker
+SmartPrompt.define_worker :multi_image_analyzer do
+  use "qwen_vl"
+  model "Qwen/Qwen2.5-VL-7B-Instruct"
+
+  content = [{ type: "text", text: params[:question] }]
+  params[:image_urls].each do |image_url|
+    content << {
+      type: "image_url",
+      image_url: { url: image_url, detail: params[:detail] || "auto" }
+    }
+  end
+
+  messages = [{ role: "user", content: content }]
+
+  sys_msg("你是一个专业的图像比较助手，能够准确比较和分析多张图像的相似之处和不同之处。", params)
+  params.merge(messages: messages)
+  send_msg
+end
+
+# Document analysis worker (for images containing text)
+SmartPrompt.define_worker :document_analyzer do
+  use "qwen_vl"
+  model "Qwen/Qwen2.5-VL-7B-Instruct"
+
+  messages = [
+    {
+      role: "user",
+      content: [
+        { type: "text", text: params[:question] || "请提取并分析这张图片中的文字内容" },
+        { type: "image_url", image_url: { url: params[:image_url], detail: "high" } }
+      ]
+    }
+  ]
+
+  sys_msg("你是一个专业的文档分析助手，能够准确提取和分析图像中的文字内容。", params)
+  params.merge(messages: messages)
+  send_msg
+end
+
+# Scene description worker
+SmartPrompt.define_worker :scene_describer do
+  use "qwen_vl"
+  model "Qwen/Qwen2.5-VL-7B-Instruct"
+
+  messages = [
+    {
+      role: "user",
+      content: [
+        { type: "text", text: "请详细描述这个场景，包括环境、人物、动作和情感氛围" },
+        { type: "image_url", image_url: { url: params[:image_url], detail: "high" } }
+      ]
+    }
+  ]
+
+  sys_msg("你是一个专业的场景描述助手，能够生动详细地描述图像中的场景。", params)
+  params.merge(messages: messages)
+  send_msg
+end

data/workers/sensenova_workers.rb

@@ -0,0 +1,62 @@
+# SenseNova workers for SmartPrompt
+#
+# These workers exercise the four SenseNova model categories through the unified
+# SenseNovaAdapter. Reuses the standard DSL: `use`, `model`, `sys_msg`, `prompt`,
+# `send_msg`, plus the multimodal/embedding/image helpers on Conversation.
+#
+# Note: `send_msg` transparently becomes streaming when the engine invokes the worker
+# via call_worker_by_stream — so :sensenova_chat serves both sync and stream callers.
+
+# 1. 商量 文本对话 (chat). Works for both call_worker and call_worker_by_stream.
+SmartPrompt.define_worker :sensenova_chat do
+  use "sensechat"
+  model params[:model] if params[:model]
+  sys_msg(params[:system] || "你是一个有帮助的中文助手，回答简洁准确。", params)
+  prompt(params[:prompt] || "你好，请介绍一下你自己。")
+  send_msg
+end
+
+# 2. 商量 图文多模态 (vision). Accepts a single image_url or image_urls (array).
+#    The multimodal content array is added to the conversation directly so send_msg
+#    sends it (send_msg reads @messages, not params).
+SmartPrompt.define_worker :sensenova_vision do
+  use "sensevision"
+  model params[:model] if params[:model]
+  sys_msg("你是一个专业的多模态图像分析助手，能够准确描述和分析图像内容。", params)
+
+  content = [{ type: "text", text: params[:question] || "请描述这张图片中的内容。" }]
+  images = params[:image_urls] || [params[:image_url]]
+  images.each do |url|
+    content << { type: "image_url", image_url: { url: url } }
+  end
+  add_message({ role: "user", content: content })
+
+  send_msg
+end
+
+# 3. Cupido 向量模型 (embeddings). Returns a normalized numeric vector of the user text.
+SmartPrompt.define_worker :sensenova_embed do
+  use "senseembedding"
+  model params[:model] if params[:model]
+  prompt(params[:text] || "商汤科技日日新大模型")
+  embeddings(params[:length] || 1024)
+end
+
+# 4. 秒画 文生图 (text-to-image). Returns the generated image(s); optionally saves to disk.
+SmartPrompt.define_worker :sensenova_image do
+  use "senseimage"
+  model params[:model] if params[:model]
+
+  images = generate_image(params[:prompt], params)
+
+  if params[:save_to_file]
+    saved = save_image(
+      images,
+      params[:output_dir] || "./generated_images",
+      params[:filename_prefix] || "sensenova"
+    )
+    { images: images, saved_files: saved }
+  else
+    images
+  end
+end