prescient 0.1.0 → 0.3.0

data/README.md

@@ -1,6 +1,10 @@
 # Prescient
 
-Prescient provides a unified interface for AI providers including Ollama (local), Anthropic Claude, OpenAI GPT, and HuggingFace models. Built for prescient applications that need AI predictions with provider switching, error handling, and fallback mechanisms.
+Prescient is a boring AI provider abstraction for Ruby. Configure your AI providers once, then use the same interface regardless of whether the request is handled by OpenAI, Anthropic, Ollama, or Hugging Face. Prescient handles provider selection, retries, health checks, and fallback.
+
+For focused guidance, see the [examples guide](examples/README.md),
+[Rails integration guide](INTEGRATION_GUIDE.md), and
+[pgvector guide](VECTOR_SEARCH_GUIDE.md).
 
 ## Features
 
@@ -16,19 +20,19 @@ Prescient provides a unified interface for AI providers including Ollama (local)
 
 ### Ollama (Local)
 
-- **Models**: Any Ollama-compatible model (llama3.1, nomic-embed-text, etc.)
+- **Models**: Any Ollama-compatible model (llama3.2, nomic-embed-text, etc.)
 - **Capabilities**: Embeddings, Text Generation, Model Management
 - **Use Case**: Privacy-focused, local deployments
 
 ### Anthropic Claude
 
-- **Models**: Claude 3 (Haiku, Sonnet, Opus)
+- **Models**: Current Claude models selected through the Anthropic Models API
 - **Capabilities**: Text Generation only (no embeddings)
 - **Use Case**: High-quality conversational AI
 
 ### OpenAI
 
-- **Models**: GPT-3.5, GPT-4, text-embedding-3-small/large
+- **Models**: Current GPT and text-embedding models selected through the OpenAI Models API
 - **Capabilities**: Embeddings, Text Generation
 - **Use Case**: Proven performance, wide model selection
 
@@ -66,21 +70,21 @@ gem install prescient
 # Ollama (Local)
 OLLAMA_URL=http://localhost:11434
 OLLAMA_EMBEDDING_MODEL=nomic-embed-text
-OLLAMA_CHAT_MODEL=llama3.1:8b
+OLLAMA_CHAT_MODEL=llama3.2:3b
 
 # Anthropic
 ANTHROPIC_API_KEY=your_api_key
-ANTHROPIC_MODEL=claude-3-haiku-20240307
+ANTHROPIC_MODEL=claude-sonnet-4-20250514
 
 # OpenAI
 OPENAI_API_KEY=your_api_key
 OPENAI_EMBEDDING_MODEL=text-embedding-3-small
-OPENAI_CHAT_MODEL=gpt-3.5-turbo
+OPENAI_CHAT_MODEL=gpt-4.1-mini
 
 # HuggingFace
 HUGGINGFACE_API_KEY=your_api_key
 HUGGINGFACE_EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2
-HUGGINGFACE_CHAT_MODEL=microsoft/DialoGPT-medium
+HUGGINGFACE_CHAT_MODEL=google/gemma-2-2b-it
 ```
 
 ### Programmatic Configuration
@@ -96,27 +100,74 @@ Prescient.configure do |config|
   config.retry_delay = 1.0
 
   # Add custom Ollama configuration
-  config.add_provider(:ollama, Prescient::Ollama::Provider,
+  config.add_provider(:ollama, Prescient::Provider::Ollama,
     url: 'http://localhost:11434',
     embedding_model: 'nomic-embed-text',
-    chat_model: 'llama3.1:8b'
+    chat_model: 'llama3.2:3b'
   )
 
   # Add Anthropic
-  config.add_provider(:anthropic, Prescient::Anthropic::Provider,
+  config.add_provider(:anthropic, Prescient::Provider::Anthropic,
     api_key: ENV['ANTHROPIC_API_KEY'],
-    model: 'claude-3-haiku-20240307'
+    model: 'claude-sonnet-4-20250514'
   )
 
   # Add OpenAI
-  config.add_provider(:openai, Prescient::OpenAI::Provider,
+  config.add_provider(:openai, Prescient::Provider::OpenAI,
+    api_key: ENV['OPENAI_API_KEY'],
+    embedding_model: 'text-embedding-3-small',
+    chat_model: 'gpt-4.1-mini'
+  )
+end
+```
+
+### Provider Fallback Configuration
+
+Prescient supports automatic fallback to backup providers when the primary provider fails. This ensures high availability for your AI applications.
+
+```ruby
+Prescient.configure do |config|
+  # Configure primary provider
+  config.add_provider(:primary, Prescient::Provider::OpenAI,
     api_key: ENV['OPENAI_API_KEY'],
     embedding_model: 'text-embedding-3-small',
-    chat_model: 'gpt-3.5-turbo'
+    chat_model: 'gpt-4.1-mini'
   )
+  
+  # Configure backup providers
+  config.add_provider(:backup1, Prescient::Provider::Anthropic,
+    api_key: ENV['ANTHROPIC_API_KEY'],
+    model: 'claude-sonnet-4-20250514'
+  )
+  
+  config.add_provider(:backup2, Prescient::Provider::Ollama,
+    url: 'http://localhost:11434',
+    embedding_model: 'nomic-embed-text',
+    chat_model: 'llama3.2:3b'
+  )
+  
+  # Configure fallback order
+  config.fallback_providers = [:backup1, :backup2]
 end
+
+# Client with fallback enabled (default)
+client = Prescient::Client.new(:primary, enable_fallback: true)
+
+# Client without fallback
+client_no_fallback = Prescient::Client.new(:primary, enable_fallback: false)
+
+# Convenience methods also support fallback
+response = Prescient.generate_response("Hello", provider: :primary, enable_fallback: true)
 ```
 
+**Fallback Behavior:**
+- When a provider fails with a persistent error, Prescient automatically tries the next available provider
+- Configured fallback providers are tried in order; the provider operation determines availability
+- If no fallback providers are configured, all configured providers are tried as fallbacks
+- Transient errors (rate limits, timeouts) still use retry logic before fallback
+- Provider-service failures, connection failures, rate limits, and unavailable models may trigger fallback; authentication and invalid-request errors are returned to the caller
+- The fallback process preserves all method arguments and options
+
 ## Usage
 
 ### Quick Start
@@ -129,7 +180,7 @@ client = Prescient.client
 
 # Generate embeddings
 embedding = client.generate_embedding("Your text here")
-# => [0.1, 0.2, 0.3, ...] (768-dimensional vector)
+# => [0.1, 0.2, 0.3, ...] (model-dependent vector dimensions)
 
 # Generate text responses
 response = client.generate_response("What is Ruby?")
@@ -170,8 +221,8 @@ response = Prescient.generate_response(query, context_items,
 )
 
 puts response[:response]
-puts "Model: #{response[:model]}"
-puts "Provider: #{response[:provider]}"
+puts "Model: " + response[:model]
+puts "Provider: " + response[:provider]
 ```
 
 ### Error Handling
@@ -192,12 +243,20 @@ end
 
 ### Health Monitoring
 
+Health results separate transport reachability from configured-model readiness:
+`reachable: true` means the provider answered, while `ready: true` means the
+configured operation models were found or validated. Fallback uses the actual
+operation and does not perform an additional health request.
+
 ```ruby
 # Check all providers
-[:ollama, :anthropic, :openai, :huggingface].each do |provider|
+Prescient.configuration.providers.keys.each do |provider|
   health = Prescient.health_check(provider: provider)
   puts "#{provider}: #{health[:status]}"
-  puts "Ready: #{health[:ready]}" if health[:ready]
+  puts "Reachable: #{health[:reachable]}"
+  puts "Ready: #{health[:ready]}"
+rescue Prescient::Error => e
+  puts "#{provider}: unavailable (#{e.message})"
 end
 ```
 
@@ -210,18 +269,18 @@ Prescient.configure do |config|
   config.add_provider(:customer_service, Prescient::Provider::OpenAI,
     api_key: ENV['OPENAI_API_KEY'],
     embedding_model: 'text-embedding-3-small',
-    chat_model: 'gpt-3.5-turbo',
+    chat_model: 'gpt-4.1-mini',
     prompt_templates: {
       system_prompt: 'You are a friendly customer service representative.',
       no_context_template: <<~TEMPLATE.strip,
-        %{system_prompt}
+        %{ system_prompt }
 
         Customer Question: %{query}
 
         Please provide a helpful response.
       TEMPLATE
       with_context_template: <<~TEMPLATE.strip
-        %{system_prompt} Use the company info below to help answer.
+        %{ system_prompt } Use the company info below to help answer.
 
         Company Information:
         %{context}
@@ -259,6 +318,7 @@ prompt_templates: {
   system_prompt: 'You are a technical documentation assistant. Provide detailed explanations with code examples.',
   # ... templates
 }
+
 ```
 
 #### Creative Writing
@@ -316,6 +376,14 @@ response = client.generate_response("I need a laptop for work", products)
 - **fields** - Array of field names available for this context type
 - **format** - Template string for displaying context items
 - **embedding_fields** - Specific fields to use when generating embeddings
+- **context_excluded_fields** - Additional field names excluded from generic embedding text; built-in metadata exclusions remain active
+
+```ruby
+config.add_provider(:openai, Prescient::Provider::OpenAI,
+  api_key: ENV['OPENAI_API_KEY'],
+  context_excluded_fields: %w[tenant_id internal_notes]
+)
+```
 
 ### Automatic Context Detection
 
@@ -337,7 +405,7 @@ The system works perfectly without any context configuration - it will:
 
 ```ruby
 # No context_configs needed - works with any data!
-client = Prescient.client(:default)
+client = Prescient.client
 response = client.generate_response("Analyze this", [
   { 'title' => 'Issue', 'content' => 'Server down', 'created_at' => '2024-01-01' },
   { 'name' => 'Alert', 'message' => 'High CPU usage', 'timestamp' => 1234567 }
@@ -346,9 +414,44 @@ response = client.generate_response("Analyze this", [
 
 See `examples/custom_contexts.rb` for complete examples.
 
+### Sensitive Provider Options
+
+`provider_info` always removes the built-in sensitive keys (`api_key`,
+`password`, `token`, and `secret`). Add project-specific keys globally with
+`sensitive_keys`; nested hashes and arrays are sanitized recursively:
+
+```ruby
+Prescient.configure do |config|
+  config.sensitive_keys = %w[workspace_secret private_key]
+end
+```
+
 ## Vector Database Integration (pgvector)
 
-Prescient integrates seamlessly with PostgreSQL's pgvector extension for storing and searching embeddings:
+`Prescient::Pgvector::Store` is an opt-in PostgreSQL integration for storing
+and searching embeddings. It accepts a PG-compatible connection, so `pg` stays
+an application dependency rather than a required dependency of this gem.
+
+```ruby
+require 'pg'
+
+connection = PG.connect(dbname: 'my_app')
+store = Prescient::Pgvector::Store.new(connection: connection, dimensions: 1536)
+store.install!
+store.create_index!
+
+embedding = Prescient.generate_embedding('Semantic search', provider: :openai)
+store.upsert(
+  id: 'document-42', embedding:, provider: 'openai', model: 'text-embedding-3-small',
+  content: 'Semantic search', metadata: { category: 'guide' }
+)
+
+results = store.search(embedding:, provider: :openai, model: 'text-embedding-3-small')
+```
+
+Every vector must exactly match the store's configured dimensions; Prescient
+never pads or truncates vectors. The existing application-schema example below
+remains available for projects that need documents, chunks, and custom metadata.
 
 ### Setup with Docker
 
@@ -356,7 +459,7 @@ The included `docker-compose.yml` provides a complete setup with PostgreSQL + pg
 
 ```bash
 # Start PostgreSQL with pgvector
-docker-compose up -d postgres
+docker compose up -d postgres
 
 # The database will automatically:
 # - Install pgvector extension
@@ -400,7 +503,7 @@ embedding = client.generate_embedding(text)
 vector_str = "[#{embedding.join(',')}]"
 db.exec_params(
   "INSERT INTO document_embeddings (document_id, embedding_provider, embedding_model, embedding_dimensions, embedding, embedding_text) VALUES ($1, $2, $3, $4, $5, $6)",
-  [doc_id, 'ollama', 'nomic-embed-text', 768, vector_str, text]
+  [doc_id, 'ollama', 'nomic-embed-text', embedding.length, vector_str, text]
 )
 
 # Perform similarity search
@@ -409,10 +512,10 @@ query_embedding = client.generate_embedding(query_text)
 query_vector = "[#{query_embedding.join(',')}]"
 
 results = db.exec_params(
-  "SELECT d.title, d.content, de.embedding <=> $1::vector AS distance 
-   FROM documents d 
-   JOIN document_embeddings de ON d.id = de.document_id 
-   ORDER BY de.embedding <=> $1::vector 
+  "SELECT d.title, d.content, de.embedding <=> $1::vector AS distance
+   FROM documents d
+   JOIN document_embeddings de ON d.id = de.document_id
+   ORDER BY de.embedding <=> $1::vector
    LIMIT 5",
   [query_vector]
 )
@@ -423,14 +526,14 @@ results = db.exec_params(
 pgvector supports three distance functions:
 
 - **Cosine Distance** (`<=>`): Best for normalized embeddings
-- **L2 Distance** (`<->`): Euclidean distance, good general purpose  
+- **L2 Distance** (`<->`): Euclidean distance, good general purpose
 - **Inner Product** (`<#>`): Dot product, useful for specific cases
 
 ```sql
 -- Cosine similarity (most common)
 ORDER BY embedding <=> query_vector
 
--- L2 distance 
+-- L2 distance
 ORDER BY embedding <-> query_vector
 
 -- Inner product
@@ -443,8 +546,8 @@ The setup automatically creates HNSW indexes for fast similarity search:
 
 ```sql
 -- Example index for cosine distance
-CREATE INDEX idx_embeddings_cosine 
-ON document_embeddings 
+CREATE INDEX idx_embeddings_cosine
+ON document_embeddings
 USING hnsw (embedding vector_cosine_ops)
 WITH (m = 16, ef_construction = 64);
 ```
@@ -457,22 +560,22 @@ Combine vector similarity with metadata filtering:
 # Search with tag filtering
 results = db.exec_params(
   "SELECT d.title, de.embedding <=> $1::vector as distance
-   FROM documents d 
+   FROM documents d
    JOIN document_embeddings de ON d.id = de.document_id
    WHERE d.metadata->'tags' ? 'programming'
-   ORDER BY de.embedding <=> $1::vector 
+   ORDER BY de.embedding <=> $1::vector
    LIMIT 5",
   [query_vector]
 )
 
-# Search with difficulty and tag filters  
+# Search with difficulty and tag filters
 results = db.exec_params(
   "SELECT d.title, de.embedding <=> $1::vector as distance
-   FROM documents d 
+   FROM documents d
    JOIN document_embeddings de ON d.id = de.document_id
    WHERE d.metadata->>'difficulty' = 'beginner'
      AND d.metadata->'tags' ?| $2::text[]
-   ORDER BY de.embedding <=> $1::vector 
+   ORDER BY de.embedding <=> $1::vector
    LIMIT 5",
   [query_vector, ['ruby', 'programming']]
 )
@@ -488,7 +591,7 @@ For large datasets, tune HNSW parameters:
 -- High accuracy (slower build, more memory)
 WITH (m = 32, ef_construction = 128)
 
--- Fast build (lower accuracy, less memory)  
+-- Fast build (lower accuracy, less memory)
 WITH (m = 8, ef_construction = 32)
 
 -- Balanced (recommended default)
@@ -502,9 +605,9 @@ WITH (m = 16, ef_construction = 64)
 SET hnsw.ef_search = 100;  -- Higher = more accurate, slower
 
 -- Use EXPLAIN ANALYZE to optimize queries
-EXPLAIN ANALYZE 
-SELECT * FROM document_embeddings 
-ORDER BY embedding <=> '[0.1,0.2,...]'::vector 
+EXPLAIN ANALYZE
+SELECT * FROM document_embeddings
+ORDER BY embedding <=> '[0.1,0.2,...]'::vector
 LIMIT 10;
 ```
 
@@ -516,14 +619,14 @@ For large documents, use chunking for better search granularity:
 def chunk_document(text, chunk_size: 500, overlap: 50)
   chunks = []
   start = 0
-  
+
   while start < text.length
     end_pos = [start + chunk_size, text.length].min
     chunk = text[start...end_pos]
     chunks << chunk
     start += chunk_size - overlap
   end
-  
+
   chunks
 end
 
@@ -541,13 +644,14 @@ Run the complete vector search example:
 
 ```bash
 # Start services
-docker-compose up -d postgres ollama
+docker compose up -d postgres ollama
 
 # Run example
 DB_HOST=localhost ruby examples/vector_search.rb
 ```
 
 The example demonstrates:
+
 - Document embedding generation and storage
 - Similarity search with different distance functions
 - Metadata filtering and advanced queries
@@ -558,7 +662,7 @@ The example demonstrates:
 ### Custom Provider Implementation
 
 ```ruby
-class MyCustomProvider < Prescient::BaseProvider
+class MyCustomProvider < Prescient::Base
   def generate_embedding(text, **options)
     # Your implementation
   end
@@ -594,16 +698,16 @@ client = Prescient.client(:ollama)
 info = client.provider_info
 
 puts info[:name]      # => :ollama
-puts info[:class]     # => "Prescient::Ollama::Provider"
+puts info[:class]     # => "Ollama"
 puts info[:available] # => true
-puts info[:options]   # => {...} (excluding sensitive data)
+puts info[:options]   # => { ... } (excluding sensitive data)
 ```
 
 ## Provider-Specific Features
 
 ### Ollama
 
-- Model management: `pull_model`, `list_models`
+- Model management: `pull_model`, `available_models`
 - Local deployment support
 - No API costs
 
@@ -617,6 +721,7 @@ puts info[:options]   # => {...} (excluding sensitive data)
 - Multiple embedding model sizes
 - Latest GPT models
 - Reliable performance
+- Uses the Chat Completions endpoint for the stable normalized response contract; the newer Responses API remains a future compatibility extension.
 
 ### HuggingFace
 
@@ -633,6 +738,7 @@ The easiest way to get started with Prescient and Ollama is using Docker Compose
 Before starting, ensure your system meets the minimum requirements for running Ollama:
 
 #### **Minimum Requirements:**
+
 - **CPU**: 4+ cores (x86_64 or ARM64)
 - **RAM**: 8GB+ (16GB recommended)
 - **Storage**: 10GB+ free space for models
@@ -640,48 +746,53 @@ Before starting, ensure your system meets the minimum requirements for running O
 
 #### **Model-Specific Requirements:**
 
-| Model | RAM Required | Storage | Notes |
-|-------|-------------|---------|-------|
-| `nomic-embed-text` | 1GB | 274MB | Embedding model |
-| `llama3.1:8b` | 8GB | 4.7GB | Chat model (8B parameters) |
-| `llama3.1:70b` | 64GB+ | 40GB | Large chat model (70B parameters) |
-| `codellama:7b` | 8GB | 3.8GB | Code generation model |
+| Model              | RAM Required | Storage | Notes                             |
+| ------------------ | ------------ | ------- | --------------------------------- |
+| `nomic-embed-text` | 1GB          | 274MB   | Embedding model                   |
+| `llama3.2:3b`      | 2GB          | 2.0GB   | Chat model (3B parameters)        |
+| `llama3.1:70b`     | 64GB+        | 40GB    | Large chat model (70B parameters) |
+| `codellama:7b`     | 8GB          | 3.8GB   | Code generation model             |
 
 #### **Performance Recommendations:**
+
 - **SSD Storage**: Significantly faster model loading
 - **GPU (Optional)**: NVIDIA GPU with 8GB+ VRAM for acceleration
 - **Network**: Stable internet for initial model downloads
 - **Docker**: 4GB+ memory limit configured
 
 #### **GPU Acceleration (Optional):**
+
 - **NVIDIA GPU**: RTX 3060+ with 8GB+ VRAM recommended
 - **CUDA**: Version 11.8+ required
 - **Docker**: NVIDIA Container Toolkit installed
 - **Performance**: 3-10x faster inference with compatible models
 
-> **💡 Tip**: Start with smaller models like `llama3.1:8b` and upgrade based on your hardware capabilities and performance needs.
+> **💡 Tip**: Start with smaller models like `llama3.2:3b` and upgrade based on your hardware capabilities and performance needs.
 
 ### Quick Start with Docker
 
 1. **Start Ollama service:**
+
    ```bash
-   docker-compose up -d ollama
+   docker compose up -d ollama
    ```
 
 2. **Pull required models:**
+
    ```bash
    # Automatic setup
-   docker-compose up ollama-init
-   
+   docker compose run --rm ollama-init
+
    # Or manual setup
    ./scripts/setup-ollama-models.sh
    ```
 
 3. **Run examples:**
+
    ```bash
    # Set environment variable
    export OLLAMA_URL=http://localhost:11434
-   
+
    # Run examples
    ruby examples/custom_contexts.rb
    ```
@@ -702,9 +813,9 @@ The included `docker-compose.yml` provides:
 services:
   ollama:
     ports:
-      - "11434:11434"  # Ollama API port
+      - "11434:11434" # Ollama API port
     volumes:
-      - ollama_data:/root/.ollama  # Persist models
+      - ollama_data:/root/.ollama # Persist models
     environment:
       - OLLAMA_HOST=0.0.0.0
       - OLLAMA_ORIGINS=*
@@ -732,7 +843,7 @@ services:
 # Ollama Configuration
 OLLAMA_URL=http://localhost:11434
 OLLAMA_EMBEDDING_MODEL=nomic-embed-text
-OLLAMA_CHAT_MODEL=llama3.1:8b
+OLLAMA_CHAT_MODEL=llama3.2:3b
 
 # Optional: Other AI providers
 OPENAI_API_KEY=your_key_here
@@ -749,7 +860,7 @@ curl http://localhost:11434/api/tags
 # Pull a specific model
 curl -X POST http://localhost:11434/api/pull \
   -H "Content-Type: application/json" \
-  -d '{"name": "llama3.1:8b"}'
+  -d '{ "name": "llama3.2:3b"}'
 
 # Health check
 curl http://localhost:11434/api/version
@@ -770,6 +881,7 @@ For production use:
 #### **Common Issues:**
 
 **Out of Memory Errors:**
+
 ```bash
 # Check available memory
 free -h
@@ -778,10 +890,11 @@ free -h
 # Settings > Resources > Memory: 8GB+
 
 # Use smaller models if hardware limited
-OLLAMA_CHAT_MODEL=llama3.1:7b ruby examples/custom_contexts.rb
+OLLAMA_CHAT_MODEL=llama3.2:3b ruby examples/custom_contexts.rb
 ```
 
 **Slow Model Loading:**
+
 ```bash
 # Check disk I/O
 iostat -x 1
@@ -791,15 +904,17 @@ iostat -x 1
 ```
 
 **Model Download Failures:**
+
 ```bash
 # Check disk space
 df -h
 
 # Manually pull models with retry
-docker exec prescient-ollama ollama pull llama3.1:8b
+docker exec prescient-ollama ollama pull llama3.2:3b
 ```
 
 **GPU Not Detected:**
+
 ```bash
 # Check NVIDIA Docker runtime
 docker run --rm --gpus all nvidia/cuda:11.8-base nvidia-smi
@@ -820,7 +935,7 @@ docker logs prescient-ollama
 # Test API response time
 time curl -X POST http://localhost:11434/api/generate \
   -H "Content-Type: application/json" \
-  -d '{"model": "llama3.1:8b", "prompt": "Hello", "stream": false}'
+  -d '{ "model": "llama3.2:3b", "prompt": "Hello", "stream": false}'
 ```
 
 ## Testing
@@ -828,11 +943,44 @@ time curl -X POST http://localhost:11434/api/generate \
 The gem includes comprehensive test coverage:
 
 ```bash
-bundle exec rspec
+bundle exec rake test
 ```
 
 ## Development
 
+### Opt-in live provider smoke tests
+
+The default test suite uses mocked provider interactions. To exercise a live
+provider explicitly, set `PRESCIENT_LIVE_SMOKE=1` and select one or more
+providers with `PRESCIENT_LIVE_PROVIDERS`:
+
+```bash
+PRESCIENT_LIVE_SMOKE=1 \
+PRESCIENT_LIVE_PROVIDERS=openai \
+OPENAI_API_KEY=... \
+bundle exec ruby -Itest test/prescient/live_provider_smoke_test.rb
+```
+
+Supported provider names are `ollama`, `anthropic`, `openai`, and
+`huggingface`. The corresponding provider environment variables and model
+overrides are honored. These tests are never live unless both opt-in
+variables are set.
+
+### RBS and Steep
+
+Validate the curated core API signatures with:
+
+```bash
+bundle exec rake rbs:validate
+```
+
+Generate disposable prototypes for comparison with:
+
+```bash
+bundle exec rake rbs:prototype
+bundle exec rake rbs:diff
+```
+
 After checking out the repo, run:
 
 ```bash
@@ -856,34 +1004,3 @@ bundle exec rake install
 ## License
 
 The gem is available as open source under the terms of the [MIT License](https://opensource.org/licenses/MIT).
-
-## Roadmap
-
-### Version 0.2.0 (Planned)
-
-- **MariaDB Vector Support**: Integration with MariaDB using external vector databases
-- **Hybrid Database Architecture**: Support for MariaDB + Milvus/Qdrant combinations
-- **Vector Database Adapters**: Pluggable adapters for different vector storage backends
-- **Enhanced Chunking Strategies**: Smart document splitting with multiple algorithms
-- **Search Result Ranking**: Advanced scoring and re-ranking capabilities
-
-### Version 0.3.0 (Future)
-
-- **Streaming Responses**: Real-time response streaming for chat applications
-- **Multi-Model Ensembles**: Combine responses from multiple AI providers
-- **Advanced Analytics**: Search performance insights and usage analytics
-- **Cloud Provider Integration**: Direct support for Pinecone, Weaviate, etc.
-
-## Changelog
-
-### Version 0.1.0
-
-- Initial release
-- Support for Ollama, Anthropic, OpenAI, and HuggingFace
-- Unified interface for embeddings and text generation
-- Comprehensive error handling and retry logic
-- Health monitoring capabilities
-- PostgreSQL pgvector integration with complete Docker setup
-- Vector similarity search with multiple distance functions
-- Document chunking and metadata filtering
-- Performance optimization guides and troubleshooting