smart_prompt 0.4.4 → 0.5.0

data/README.md

@@ -11,13 +11,24 @@ SmartPrompt is a powerful Ruby gem that provides an elegant domain-specific lang
 
 ### Multi-LLM Support
 - **OpenAI API Compatible**: Full support for OpenAI GPT models and compatible APIs
-- **Llama.cpp Integration**: Direct integration with local Llama.cpp servers  
+- **Anthropic Claude**: Native support for Claude models with multimodal capabilities
+- **SenseNova (商汤日日新)**: One adapter covers chat (商量), multimodal vision (图文多模态), Cupido embeddings (向量), and 秒画 text-to-image — see `examples/sensenova_example.rb`
+- **智谱 AI (BigModel / GLM)**: One adapter covers all categories — chat (GLM-4), vision (GLM-4V), embeddings (embedding-3), text-to-image (CogView), text-to-video (CogVideoX), TTS (GLM-TTS), ASR (GLM-ASR) — see `examples/zhipu_example.rb`
+- **Llama.cpp Integration**: Direct integration with local Llama.cpp servers
 - **Extensible Adapters**: Easy-to-extend adapter system for new LLM providers
 - **Unified Interface**: Same API regardless of the underlying LLM provider
 
+### Multimodal AI Capabilities
+- **Vision Models**: Support for image understanding and analysis
+- **Image Generation**: Create images from text prompts using diffusion models
+- **Video Generation**: Generate videos from text or image prompts
+- **Text-to-Speech**: Convert text to natural-sounding speech
+- **Speech-to-Text**: Transcribe audio files to text with multi-language support
+
 ### Flexible Architecture
 - **Worker-based Tasks**: Define reusable workers for specific AI tasks
 - **Template System**: ERB-based prompt templates with parameter injection
+- **Intelligent History Management**: Session isolation, automatic compression, and multiple context strategies
 - **Conversation Management**: Built-in conversation history and context management
 - **Streaming Support**: Real-time response streaming for better user experience
 
@@ -26,6 +37,8 @@ SmartPrompt is a powerful Ruby gem that provides an elegant domain-specific lang
 - **Retry Logic**: Robust error handling with configurable retry mechanisms
 - **Embeddings**: Text embedding generation for semantic search and RAG applications
 - **Configuration-driven**: YAML-based configuration for easy deployment management
+- **Batch Processing**: Efficient processing of multiple files and tasks
+- **Language Detection**: Automatic language identification from text and audio
 
 ### Production Ready
 - **Comprehensive Logging**: Detailed logging for debugging and monitoring
@@ -61,6 +74,7 @@ Create a YAML configuration file (`config/smart_prompt.yml`):
 # Adapter definitions
 adapters:
   openai: OpenAIAdapter
+  anthropic: AnthropicAdapter
 # LLM configurations
 llms:
   SiliconFlow:
@@ -68,40 +82,25 @@ llms:
     url: https://api.siliconflow.cn/v1/
     api_key: ENV["APIKey"]
     default_model: Qwen/Qwen2.5-7B-Instruct
-  local:
+  claude:
+    adapter: anthropic
+    api_key: ENV["ANTHROPIC_API_KEY"]
+    model: claude-3-5-sonnet-20241022
+    temperature: 0.7
+    max_tokens: 4096
+  llamacpp:
     adapter: openai
     url: http://localhost:8080/    
   ollama:
     adapter: openai
     url: http://localhost:11434/
     default_model: deepseek-r1
-  gemma4_local:
-    adapter: openai
-    url: http://localhost:8000/v1
-    api_key: dummy
-    default_model: gemma-4-12B-it
-    temperature: 1.0
-    top_p: 0.95
-    top_k: 64
   deepseek:
     adapter: openai
     url: https://api.deepseek.com
     api_key: ENV["DSKEY"]
     default_model: deepseek-reasoner
 
-# Model aliases
-models:
-  local/qwen3.5:
-    use: local
-    model: qwen3.5
-  deepseekv3.2:
-    use: SiliconFlow
-    model: Pro/deepseek-ai/DeepSeek-V3.2
-  gemma4/12b:
-    use: gemma4_local
-    model: gemma-4-12B-it
-    max_tokens: 1024
-
 # Default settings
 default_llm: SiliconFlow
 template_path: "./templates"
@@ -129,8 +128,9 @@ Create worker files in your `workers/` directory:
 **workers/chat_worker.rb**:
 ```ruby
 SmartPrompt.define_worker :chat_assistant do
-  # Use a configured model alias
-  use_model "deepseekv3.2"
+  # Use a specific LLM
+  use "SiliconFlow"
+  model "deepseek-ai/DeepSeek-V3"
   # Set system message
   sys_msg("You are a helpful AI assistant.", params)
   # Use template with parameters  
@@ -182,26 +182,6 @@ engine.call_worker_by_stream(:streaming_chat, {
 end
 ```
 
-### Gemma 4 12B Multimodal
-
-Gemma 4 12B can be connected through OpenAI-compatible local servers such as LiteRT-LM, LM Studio, Ollama, or llama.cpp. SmartPrompt places images before text and audio after text to match Gemma 4 multimodal best practices.
-
-```ruby
-SmartPrompt.define_worker :gemma_multimodal_assistant do
-  use_model "gemma4/12b"
-  thinking params.fetch(:thinking, true)
-  sys_msg("You are a precise local multimodal assistant.", params)
-
-  image(params[:image], token_budget: params[:token_budget] || 280) if params[:image]
-  video(params[:video], fps: 1, max_seconds: 60) if params[:video]
-  audio(params[:audio]) if params[:audio]
-  prompt(params[:message])
-
-  request_options(response_format: { type: "json_object" }) if params[:json]
-  send_msg
-end
-```
-
 ### Tool Integration
 
 ```ruby
@@ -238,7 +218,10 @@ end
 
 ### Conversation History
 
+SmartPrompt provides intelligent conversation history management with session isolation, automatic compression, and multiple context strategies.
+
 ```ruby
+# Basic usage with automatic history management
 SmartPrompt.define_worker :conversational_chat do
   use "deepseek"
   model "deepseek-chat"
@@ -246,8 +229,38 @@ SmartPrompt.define_worker :conversational_chat do
   prompt(params[:message], with_history: true)
   send_msg
 end
+
+# Advanced usage with explicit session management
+SmartPrompt.define_worker :session_chat do
+  use "deepseek"
+  model "deepseek-chat"
+  
+  # Use session_id for isolated conversations
+  session_id = params[:session_id] || "default"
+  
+  # Configure session behavior
+  session_config = {
+    max_messages: 100,
+    max_tokens: 4000,
+    context_strategy: :sliding_window  # or :relevance_based, :summary_based, :hybrid
+  }
+  
+  sys_msg("You are a helpful assistant.", params)
+  prompt(params[:message], with_history: true)
+  params.merge(session_id: session_id, session_config: session_config)
+  send_msg
+end
 ```
 
+**History Management Features:**
+- **Session Isolation**: Each conversation has independent history
+- **Context Strategies**: Choose from sliding window, relevance-based, summary-based, or hybrid
+- **Automatic Compression**: Reduce token usage while preserving context
+- **Persistence**: Save and restore conversations across restarts
+- **Performance**: LRU caching and async I/O for optimal performance
+
+See [History Management Guide](HISTORY_MANAGEMENT_GUIDE.md) for detailed documentation.
+
 ### Embeddings Generation
 
 ```ruby
@@ -265,6 +278,78 @@ embeddings = engine.call_worker(:text_embedder, {
 })
 ```
 
+### Multimodal AI Examples
+
+#### Image Generation
+```ruby
+# Generate image from text prompt (SiliconFlow /v1/images/generations)
+result = engine.call_worker(:image_generator, {
+  prompt: "A beautiful sunset over mountains",
+  image_size: "1024x1024",   # "widthxheight"; aliases: size:
+  batch_size: 1,             # only Kolors; aliases: n:
+  negative_prompt: "blurry, low quality",
+  save_to_file: true,
+  output_dir: "./generated_images"
+})
+
+puts "Generated #{result[:images].size} image(s)"
+puts "First image URL: #{result[:images].first[:url]}"
+puts "Saved files: #{result[:saved_files]}"
+```
+
+#### Video Generation
+```ruby
+# Generate video from text prompt
+result = engine.call_worker(:video_generator, {
+  prompt: "A cat playing with a ball of yarn",
+  duration: 5,
+  resolution: "720p",
+  save_to_file: true,
+  output_dir: "./generated_videos"
+})
+
+puts "Video generation started: #{result[:video_id]}"
+puts "Check status with: engine.call_worker(:video_status, {video_id: '#{result[:video_id]}'})"
+```
+
+#### Text-to-Speech
+```ruby
+# Convert text to speech
+result = engine.call_worker(:tts_synthesizer, {
+  text: "Welcome to SmartPrompt, your AI assistant",
+  voice: "alloy",
+  speed: 1.0,
+  save_to_file: true,
+  output_dir: "./generated_audio"
+})
+
+puts "Audio file created: #{result[:audio_file][:file_path]}"
+```
+
+#### Speech-to-Text
+```ruby
+# Transcribe audio to text
+result = engine.call_worker(:stt_transcriber, {
+  audio_file: "./audio.wav",
+  language: "en",
+  response_format: "json"
+})
+
+puts "Transcribed text: #{result[:transcription][:text]}"
+puts "Language: #{result[:transcription][:language]}"
+```
+
+#### Vision Analysis
+```ruby
+# Analyze image with vision model
+result = engine.call_worker(:vision_analyzer, {
+  image_file: "./image.jpg",
+  prompt: "Describe what you see in this image"
+})
+
+puts "Analysis: #{result[:response]}"
+```
+
 ## 🏗️ Architecture Overview
 
 SmartPrompt follows a modular architecture:
@@ -282,6 +367,13 @@ SmartPrompt follows a modular architecture:
                    │Workers│ │Conv.│ │Template│
                    │       │ │Mgmt │ │ System │
                    └───────┘ └─────┘ └────────┘
+                                │
+                       ┌────────┴────────┐
+                       │                 │
+                   ┌───▼────────┐  ┌─────▼──────┐
+                   │  History   │  │Persistence │
+                   │  Manager   │  │   Layer    │
+                   └────────────┘  └────────────┘
 ```
 
 ### Core Components
@@ -289,8 +381,10 @@ SmartPrompt follows a modular architecture:
 - **Engine**: Central orchestrator managing configuration, adapters, and workers
 - **Workers**: Reusable task definitions with embedded business logic
 - **Conversation**: Context and message history management
-- **Adapters**: LLM provider integrations (OpenAI, Llama.cpp, etc.)
+- **History Manager**: Intelligent conversation history with session isolation and context strategies
+- **Adapters**: LLM provider integrations (OpenAI, Anthropic, Llama.cpp, etc.)
 - **Templates**: ERB-based prompt template system
+- **Persistence Layer**: Save and restore conversation history across restarts
 
 ## 🔧 Configuration Reference
 
@@ -298,32 +392,178 @@ SmartPrompt follows a modular architecture:
 
 ```yaml
 adapters:
-  openai: "OpenAIAdapter"      # For OpenAI API
+  openai: "OpenAIAdapter"              # For OpenAI API
+  anthropic: "AnthropicAdapter"        # For Anthropic Claude API
+  sensenova: "SenseNovaAdapter"        # For 商汤 SenseNova (chat/vision/embeddings/image)
+  zhipu: "ZhipuAIAdapter"              # For 智谱 BigModel/GLM (chat/vision/embed/image/video/tts/asr)
+  multimodal: "MultimodalAdapter"      # For vision models
+  image_generation: "ImageGenerationAdapter"    # For image generation
+  video_generation: "VideoGenerationAdapter"    # For video generation
+  tts: "TTSAdapter"                    # For text-to-speech
+  stt: "STTAdapter"                    # For speech-to-text
 ```
 
 ### LLM Configuration
 
 ```yaml
 llms:
-  model_name:
-    adapter: "adapter_name"
-    api_key: "your_api_key"     # Can use ENV['KEY_NAME']
-    url: "https://api.url"
-    model: "model_identifier"
+  # Text models
+  gpt:
+    adapter: "openai"
+    api_key: ENV["OPENAI_API_KEY"]
+    model: "gpt-4"
     temperature: 0.7
-    # Additional provider-specific options
-```
 
-### Model Alias Configuration
+  # Anthropic Claude models
+  claude:
+    adapter: "anthropic"
+    api_key: ENV["ANTHROPIC_API_KEY"]
+    model: "claude-3-5-sonnet-20241022"
+    temperature: 0.7
+    max_tokens: 4096
 
-```yaml
-models:
-  model_alias:
-    use: "llm_name"
-    model: "model_identifier"
-```
+  claude_opus:
+    adapter: "anthropic"
+    api_key: ENV["ANTHROPIC_API_KEY"]
+    model: "claude-3-opus-20240229"
+    temperature: 0.7
+    max_tokens: 4096
 
-In a worker, `use_model "model_alias"` is equivalent to calling `use "llm_name"` and `model "model_identifier"`.
+  claude_haiku:
+    adapter: "anthropic"
+    api_key: ENV["ANTHROPIC_API_KEY"]
+    model: "claude-3-5-haiku-20241022"
+    temperature: 0.7
+    max_tokens: 4096
+
+  # Custom Anthropic endpoint (for proxy or custom deployment)
+  claude_custom:
+    adapter: "anthropic"
+    api_key: ENV["ANTHROPIC_API_KEY"]
+    url: "https://your-custom-endpoint.com"
+    model: "claude-3-5-sonnet-20241022"
+    temperature: 0.7
+    max_tokens: 4096
+
+  # 商汤 SenseNova — one adapter covers all four model categories; just change `model`.
+  # Free-tier models run on token.sensenova.cn/v1; paid models (SenseChat-5, SenseNova-V6-*
+  # , Cupido) run on api.sensenova.cn/compatible-mode/v2 (returns 403 if your key lacks them).
+  sensechat:                          # 商量 文本对话 (free-tier)
+    adapter: "sensenova"
+    url: "https://token.sensenova.cn/v1"
+    api_key: ENV["SENSENOVA_API_KEY"]
+    model: "sensenova-6.7-flash-lite"
+    temperature: 0.7
+    # Optional SenseNova sampling extras (forwarded to /chat/completions):
+    # reasoning_effort: "medium"
+    # max_completion_tokens: 4096
+    # Paid: url https://api.sensenova.cn/compatible-mode/v2, model SenseChat-5
+
+  sensevision:                        # 商量 图文多模态 (flash-lite is natively multimodal)
+    adapter: "sensenova"
+    url: "https://token.sensenova.cn/v1"
+    api_key: ENV["SENSENOVA_API_KEY"]
+    model: "sensenova-6.7-flash-lite"
+    # Paid: url https://api.sensenova.cn/compatible-mode/v2, model SenseNova-V6-Pro
+
+  senseembedding:                     # Cupido 向量模型 (paid; native endpoint)
+    adapter: "sensenova"
+    url: "https://api.sensenova.cn/compatible-mode/v2"
+    embeddings_url: "https://api.sensenova.cn/v1/llm/embeddings"
+    api_key: ENV["SENSENOVA_API_KEY"]
+    model: "Cupido"
+
+  senseimage:                         # 秒画 文生图 (sensenova-u1-fast; token.sensenova.cn base)
+    adapter: "sensenova"
+    url: "https://token.sensenova.cn/v1"
+    image_url: "https://token.sensenova.cn/v1/images/generations"
+    api_key: ENV["SENSENOVA_API_KEY"]
+    model: "sensenova-u1-fast"
+    # sensenova-u1-fast only accepts specific sizes (default 2048x2048); see
+    # VALID_IMAGE_SIZES in sensenova_adapter.rb.
+
+  # 智谱 AI (BigModel/GLM) — one adapter covers all categories; just change `model`.
+  # Base https://open.bigmodel.cn/api/paas/v4 ; Bearer auth. Defaults use free-tier models.
+  glm:                                # 文本对话 (free glm-4-flash; paid glm-4-plus/glm-5.2)
+    adapter: "zhipu"
+    url: "https://open.bigmodel.cn/api/paas/v4"
+    api_key: ENV["ZHIPUAI_API_KEY"]
+    model: "glm-4-flash"
+    temperature: 0.7
+    # CodeGeeX-4: set `coding: true` and model: codegeex-4 (uses the coding base)
+
+  glm_vision:                         # 图文多模态 (free glm-4v-flash; paid glm-4v-plus)
+    adapter: "zhipu"
+    url: "https://open.bigmodel.cn/api/paas/v4"
+    api_key: ENV["ZHIPUAI_API_KEY"]
+    model: "glm-4v-flash"
+
+  embedding:                          # 向量模型 (embedding-3; custom dimensions 256/512/1024/2048)
+    adapter: "zhipu"
+    url: "https://open.bigmodel.cn/api/paas/v4"
+    api_key: ENV["ZHIPUAI_API_KEY"]
+    model: "embedding-3"
+    dimensions: 1024
+
+  cogview:                            # 文生图 (free cogview-3-flash; paid cogview-4/glm-image)
+    adapter: "zhipu"
+    url: "https://open.bigmodel.cn/api/paas/v4"
+    api_key: ENV["ZHIPUAI_API_KEY"]
+    model: "cogview-3-flash"
+
+  cogvideo:                           # 文生视频 (async submit->poll->download; free cogvideox-flash)
+    adapter: "zhipu"
+    url: "https://open.bigmodel.cn/api/paas/v4"
+    api_key: ENV["ZHIPUAI_API_KEY"]
+    model: "cogvideox-flash"
+
+  glm_tts:                            # 语音合成 (GLM-TTS)
+    adapter: "zhipu"
+    url: "https://open.bigmodel.cn/api/paas/v4"
+    api_key: ENV["ZHIPUAI_API_KEY"]
+    model: "glm-tts"
+
+  glm_asr:                            # 语音识别 (GLM-ASR-2512)
+    adapter: "zhipu"
+    url: "https://open.bigmodel.cn/api/paas/v4"
+    api_key: ENV["ZHIPUAI_API_KEY"]
+    model: "glm-asr-2512"
+
+  # Vision models
+  vision:
+    adapter: "multimodal"
+    url: "https://api.siliconflow.cn/v1/"
+    api_key: ENV["SILICONFLOW_API_KEY"]
+    model: "Qwen/Qwen2.5-VL-7B-Instruct"
+
+  # Image generation (Kolors supports batch_size/guidance_scale; see Qwen-Image for cfg)
+  image_gen:
+    adapter: "image_generation"
+    url: "https://api.siliconflow.cn/v1/"
+    api_key: ENV["SILICONFLOW_API_KEY"]
+    model: "Kwai-Kolors/Kolors"
+
+  # Video generation
+  video_gen:
+    adapter: "video_generation"
+    url: "https://api.siliconflow.cn/v1/"
+    api_key: ENV["SILICONFLOW_API_KEY"]
+    model: "Wan-AI/Wan2.2-T2V-A14B"
+
+  # Text-to-speech
+  tts_service:
+    adapter: "tts"
+    url: "https://api.siliconflow.cn/v1/"
+    api_key: ENV["SILICONFLOW_API_KEY"]
+    model: "FunAudioLLM/CosyVoice2-0.5B"
+
+  # Speech-to-text
+  stt_service:
+    adapter: "stt"
+    url: "https://api.siliconflow.cn/v1/"
+    api_key: ENV["SILICONFLOW_API_KEY"]
+    model: "FunAudioLLM/CosyVoice2-0.5B"
+```
 
 ### Path Configuration
 
@@ -398,20 +638,27 @@ end
 ## 🚀 Real-world Use Cases
 
 - **Chatbots and Conversational AI**: Build sophisticated chatbots with context awareness
-- **Content Generation**: Automated content creation with template-driven prompts  
+- **Content Generation**: Automated content creation with template-driven prompts
 - **Code Analysis**: AI-powered code review and documentation generation
 - **Customer Support**: Intelligent ticket routing and response suggestions
 - **Data Processing**: LLM-powered data extraction and transformation
 - **Educational Tools**: AI tutors and learning assistance systems
+- **Multimedia Content Creation**: Generate images, videos, and audio content
+- **Voice Interfaces**: Build voice-enabled applications with TTS and STT
+- **Visual Analysis**: Image understanding and object detection applications
+- **Accessibility Tools**: Audio descriptions, text-to-speech for visually impaired
 
 ## 🛣️ Roadmap
 
+- [x] **Multimodal AI Support** - Vision, Image Generation, Video Generation, TTS, STT
 - [ ] Additional LLM provider adapters (Anthropic Claude, Google PaLM)
 - [ ] Visual prompt builder and management interface
 - [ ] Enhanced caching and performance optimizations
 - [ ] Integration with vector databases for RAG applications
 - [ ] Built-in evaluation and testing framework for prompts
 - [ ] Distributed worker execution support
+- [ ] Real-time audio/video streaming support
+- [ ] Advanced multimodal prompt chaining
 
 ## 🤝 Contributing
 
@@ -442,4 +689,4 @@ This project is licensed under the MIT License - see the [LICENSE.txt](LICENSE.t
 
 ---
 
-**SmartPrompt** - Making LLM integration in Ruby applications simple, powerful, and elegant.
+**SmartPrompt** - Making LLM integration in Ruby applications simple, powerful, and elegant.

data/Rakefile

@@ -1,4 +1,13 @@
 # frozen_string_literal: true
 
 require "bundler/gem_tasks"
-task default: %i[]
+require "rake/testtask"
+
+Rake::TestTask.new(:test) do |t|
+  t.libs << "lib"
+  t.libs << "test"
+  t.test_files = FileList["test/**/*_test.rb"]
+  t.verbose = true
+end
+
+task default: :test

data/config/anthropic_config.yml

@@ -0,0 +1,151 @@
+# Anthropic Configuration for SmartPrompt
+# This configuration enables Anthropic Claude models
+
+# Adapter definitions
+adapters:
+  openai: "OpenAIAdapter"
+  anthropic: "AnthropicAdapter"
+
+# LLM configurations
+llms:
+  deepseek_anthropic:
+    adapter: anthropic
+    api_key: ENV["ANTHROPIC_AUTH_TOKEN"]
+    url: "https://api.deepseek.com/anthropic"
+    temperature: 0.7
+    max_tokens: 4096
+  deepseek:
+    adapter: openai
+    api_key: ENV["DSKEY"]
+    url: "https://api.deepseek.com"
+
+# Path configurations
+template_path: "./templates"
+worker_path: "./workers"
+logger_file: "./logs/smart_prompt.log"
+
+# Advanced settings
+advanced:
+  # Timeout settings (in seconds)
+  request_timeout: 240
+  connection_timeout: 30
+
+  # Retry settings
+  max_retries: 3
+  retry_delay: 2
+
+  # Rate limiting
+  requests_per_minute: 60
+
+# History Management Configuration
+# SmartPrompt provides intelligent conversation history management with session isolation,
+# automatic compression, and multiple context strategies.
+history:
+  # Cache Configuration
+  # Maximum number of sessions to keep in memory (LRU eviction)
+  cache_size: 100
+
+  # Default Session Configuration
+  # These settings apply to all sessions unless overridden
+  session_defaults:
+    max_messages: 100              # Maximum messages per session (older messages removed)
+    max_tokens: 4000               # Maximum tokens per session (enforced during context retrieval)
+    context_strategy: sliding_window  # Default strategy: sliding_window, relevance_based, summary_based, hybrid
+    preserve_system_messages: true # Always keep system messages regardless of limits
+
+  # Context Strategy Configurations
+  # Each strategy has specific parameters for fine-tuning behavior
+  strategies:
+    # Sliding Window: Keep the most recent N messages
+    sliding_window:
+      window_size: 10              # Number of recent messages to keep
+      preserve_system: true        # Always include system messages
+
+    # Relevance-Based: Select messages based on semantic similarity
+    relevance_based:
+      top_k: 10                    # Number of most relevant messages to select
+      recency_weight: 0.3          # Weight for recency (0.0-1.0)
+      relevance_weight: 0.7        # Weight for relevance (0.0-1.0)
+      embedding_service: null      # Optional: embedding service for semantic similarity
+
+    # Summary-Based: Automatically compress old messages into summaries
+    summary_based:
+      summary_threshold: 20        # Trigger summarization after this many messages
+      keep_recent: 5               # Number of recent messages to keep uncompressed
+      compression_ratio: 0.5       # Target compression ratio (0.0-1.0)
+
+    # Hybrid: Adaptively combine multiple strategies
+    hybrid:
+      mode: adaptive               # Mode: 'adaptive' (auto-select) or 'combined' (merge results)
+      sliding_window: {}           # Override sliding window config
+      relevance_based: {}          # Override relevance-based config
+      summary_based: {}            # Override summary-based config
+
+  # Compression Configuration
+  # Automatic summarization to reduce token usage
+  compression:
+    enabled: true                  # Enable automatic compression
+    auto_compress_threshold: 50    # Auto-compress when session exceeds this many messages
+    compression_ratio: 0.5         # Target compression ratio
+    llm_adapter: null              # LLM to use for summarization (uses default if null)
+
+  # Persistence Configuration
+  # Save and restore conversation history across restarts
+  persistence:
+    enabled: true                  # Enable persistence to disk
+    backend: filesystem            # Backend type: 'filesystem' (more backends coming soon)
+    storage_path: "./history_data" # Directory for storing session data
+    async: true                    # Use async writes for better performance
+
+  # Cleanup Configuration
+  # Automatic cleanup of old or expired sessions
+  cleanup:
+    auto_cleanup: false            # Enable automatic cleanup thread
+    cleanup_interval: 3600         # Cleanup interval in seconds (1 hour)
+    session_ttl: 86400             # Session time-to-live in seconds (24 hours)
+    cleanup_callback: null         # Optional: custom cleanup logic (Ruby proc)
+
+  # Monitoring Configuration
+  # Logging and metrics for debugging and monitoring
+  monitoring:
+    enabled: true                  # Enable monitoring and logging
+    log_level: info                # Log level: debug, info, warn, error
+    metrics_format: prometheus     # Metrics format: prometheus, json, hash
+
+# Example Configurations for Different Use Cases:
+#
+# 1. High-Volume Chat Application (optimize for performance):
+#    cache_size: 1000
+#    session_defaults:
+#      max_messages: 50
+#      max_tokens: 2000
+#      context_strategy: sliding_window
+#    cleanup:
+#      auto_cleanup: true
+#      session_ttl: 3600  # 1 hour
+#
+# 2. Long-Running Conversations (optimize for context retention):
+#    session_defaults:
+#      max_messages: 500
+#      max_tokens: 16000
+#      context_strategy: summary_based
+#    compression:
+#      enabled: true
+#      auto_compress_threshold: 100
+#
+# 3. Semantic Search Application (optimize for relevance):
+#    session_defaults:
+#      context_strategy: relevance_based
+#    strategies:
+#      relevance_based:
+#        top_k: 20
+#        recency_weight: 0.2
+#        relevance_weight: 0.8
+#
+# 4. Development/Testing (disable persistence and cleanup):
+#    persistence:
+#      enabled: false
+#    cleanup:
+#      auto_cleanup: false
+#    monitoring:
+#      log_level: debug