ruby_llm 1.6.2 → 1.6.4

data/lib/tasks/models.rake

@@ -0,0 +1,514 @@
+# frozen_string_literal: true
+
+require 'dotenv/load'
+require 'ruby_llm'
+require 'json'
+require 'json-schema'
+require 'fileutils'
+
+desc 'Update models, docs, and aliases'
+task models: ['models:update', 'models:docs', 'models:aliases']
+
+namespace :models do
+  desc 'Update available models from providers (API keys needed)'
+  task :update do
+    puts 'Configuring RubyLLM...'
+    configure_from_env
+    refresh_models
+    display_model_stats
+  end
+
+  desc 'Generate available models documentation'
+  task :docs do
+    FileUtils.mkdir_p('docs/_reference')
+    output = generate_models_markdown
+    File.write('docs/_reference/available-models.md', output)
+    puts 'Generated docs/_reference/available-models.md'
+  end
+
+  desc 'Generate model aliases from registry'
+  task :aliases do
+    generate_aliases
+  end
+end
+
+# Keep aliases:generate for backwards compatibility
+namespace :aliases do
+  task generate: ['models:aliases']
+end
+
+def configure_from_env
+  RubyLLM.configure do |config|
+    config.openai_api_key = ENV.fetch('OPENAI_API_KEY', nil)
+    config.anthropic_api_key = ENV.fetch('ANTHROPIC_API_KEY', nil)
+    config.gemini_api_key = ENV.fetch('GEMINI_API_KEY', nil)
+    config.deepseek_api_key = ENV.fetch('DEEPSEEK_API_KEY', nil)
+    config.perplexity_api_key = ENV.fetch('PERPLEXITY_API_KEY', nil)
+    config.openrouter_api_key = ENV.fetch('OPENROUTER_API_KEY', nil)
+    config.mistral_api_key = ENV.fetch('MISTRAL_API_KEY', nil)
+    configure_bedrock(config)
+    config.request_timeout = 30
+  end
+end
+
+def configure_bedrock(config)
+  config.bedrock_api_key = ENV.fetch('AWS_ACCESS_KEY_ID', nil)
+  config.bedrock_secret_key = ENV.fetch('AWS_SECRET_ACCESS_KEY', nil)
+  config.bedrock_region = ENV.fetch('AWS_REGION', nil)
+  config.bedrock_session_token = ENV.fetch('AWS_SESSION_TOKEN', nil)
+end
+
+def refresh_models
+  initial_count = RubyLLM.models.all.size
+  puts "Refreshing models (#{initial_count} cached)..."
+
+  models = RubyLLM.models.refresh!
+
+  if models.all.empty? && initial_count.zero?
+    puts 'Error: Failed to fetch models.'
+    exit(1)
+  elsif models.all.size == initial_count && initial_count.positive?
+    puts 'Warning: Model list unchanged.'
+  else
+    puts 'Validating models...'
+    validate_models!(models)
+
+    puts "Saving models.json (#{models.all.size} models)"
+    models.save_models
+  end
+
+  @models = models
+end
+
+def validate_models!(models)
+  schema_path = RubyLLM::Models.schema_file
+  models_data = models.all.map(&:to_h)
+
+  validation_errors = JSON::Validator.fully_validate(schema_path, models_data)
+
+  unless validation_errors.empty?
+    # Save failed models for inspection
+    failed_path = File.expand_path('../ruby_llm/models.failed.json', __dir__)
+    File.write(failed_path, JSON.pretty_generate(models_data))
+
+    puts 'ERROR: Models validation failed:'
+    puts "\nValidation errors:"
+    validation_errors.first(10).each { |error| puts "  - #{error}" }
+    puts "  ... and #{validation_errors.size - 10} more errors" if validation_errors.size > 10
+    puts "-> Failed models saved to: #{failed_path}"
+    exit(1)
+  end
+
+  puts '✓ Models validation passed'
+end
+
+def display_model_stats
+  puts "\nModel count:"
+  provider_counts = @models.all.group_by(&:provider).transform_values(&:count)
+
+  RubyLLM::Provider.providers.each do |sym, provider_class|
+    name = provider_class.name
+    count = provider_counts[sym.to_s] || 0
+    status = status(sym)
+    puts "  #{name}: #{count} models #{status}"
+  end
+
+  puts 'Refresh complete.'
+end
+
+def status(provider_sym)
+  provider_class = RubyLLM::Provider.providers[provider_sym]
+  if provider_class.local?
+    ' (LOCAL - SKIP)'
+  elsif provider_class.configured?(RubyLLM.config)
+    ' (OK)'
+  else
+    ' (NOT CONFIGURED)'
+  end
+end
+
+def generate_models_markdown
+  <<~MARKDOWN
+    ---
+    layout: default
+    title: Available Models
+    nav_order: 1
+    description: Browse hundreds of AI models from every major provider. Always up-to-date, automatically generated.
+    redirect_from:
+      - /guides/available-models
+    ---
+
+    # {{ page.title }}
+    {: .no_toc }
+
+    {{ page.description }}
+    {: .fs-6 .fw-300 }
+
+    ## Table of contents
+    {: .no_toc .text-delta }
+
+    1. TOC
+    {:toc}
+
+    ---
+
+    After reading this guide, you will know:
+
+    * How RubyLLM's model registry works and where data comes from
+    * How to find models by provider, capability, or purpose
+    * What information is available for each model
+    * How to use model aliases for simpler configuration
+
+    ## How Model Data Works
+
+    RubyLLM's model registry combines data from multiple sources:
+
+    - **OpenAI, Anthropic, DeepSeek, Gemini**: Data from [Parsera](https://api.parsera.org/v1/llm-specs)
+    - **OpenRouter**: Direct from OpenRouter's API
+    - **Other providers**: Defined in `capabilities.rb` files
+
+    ## Contributing Model Updates
+
+    **For major providers** (OpenAI, Anthropic, DeepSeek, Gemini): File issues with [Parsera](https://github.com/parsera-labs/api-llm-specs/issues) for public model data corrections.
+
+    **For other providers**: Edit `lib/ruby_llm/providers/<provider>/capabilities.rb` then run `rake models:update`.
+
+    See the [Contributing Guide](https://github.com/crmne/ruby_llm/blob/main/CONTRIBUTING.md) for details.
+
+    ## Last Updated
+    {: .d-inline-block }
+
+    #{Time.now.utc.strftime('%Y-%m-%d')}
+    {: .label .label-green }
+
+    ## Models by Provider
+
+    #{generate_provider_sections}
+
+    ## Models by Capability
+
+    #{generate_capability_sections}
+
+    ## Models by Modality
+
+    #{generate_modality_sections}
+  MARKDOWN
+end
+
+def generate_provider_sections
+  RubyLLM::Provider.providers.filter_map do |provider, provider_class|
+    models = RubyLLM.models.by_provider(provider)
+    next if models.none?
+
+    <<~PROVIDER
+      ### #{provider_class.name} (#{models.count})
+
+      #{models_table(models)}
+    PROVIDER
+  end.join("\n\n")
+end
+
+def generate_capability_sections
+  capabilities = {
+    'Function Calling' => RubyLLM.models.select(&:function_calling?),
+    'Structured Output' => RubyLLM.models.select(&:structured_output?),
+    'Streaming' => RubyLLM.models.select { |m| m.capabilities.include?('streaming') },
+    'Batch Processing' => RubyLLM.models.select { |m| m.capabilities.include?('batch') }
+  }
+
+  capabilities.filter_map do |capability, models|
+    next if models.none?
+
+    <<~CAPABILITY
+      ### #{capability} (#{models.count})
+
+      #{models_table(models)}
+    CAPABILITY
+  end.join("\n\n")
+end
+
+def generate_modality_sections # rubocop:disable Metrics/PerceivedComplexity
+  sections = []
+
+  vision_models = RubyLLM.models.select { |m| (m.modalities.input || []).include?('image') }
+  if vision_models.any?
+    sections << <<~SECTION
+      ### Vision Models (#{vision_models.count})
+
+      Models that can process images:
+
+      #{models_table(vision_models)}
+    SECTION
+  end
+
+  audio_models = RubyLLM.models.select { |m| (m.modalities.input || []).include?('audio') }
+  if audio_models.any?
+    sections << <<~SECTION
+      ### Audio Input Models (#{audio_models.count})
+
+      Models that can process audio:
+
+      #{models_table(audio_models)}
+    SECTION
+  end
+
+  pdf_models = RubyLLM.models.select { |m| (m.modalities.input || []).include?('pdf') }
+  if pdf_models.any?
+    sections << <<~SECTION
+      ### PDF Models (#{pdf_models.count})
+
+      Models that can process PDF documents:
+
+      #{models_table(pdf_models)}
+    SECTION
+  end
+
+  embedding_models = RubyLLM.models.select { |m| (m.modalities.output || []).include?('embeddings') }
+  if embedding_models.any?
+    sections << <<~SECTION
+      ### Embedding Models (#{embedding_models.count})
+
+      Models that generate embeddings:
+
+      #{models_table(embedding_models)}
+    SECTION
+  end
+
+  sections.join("\n\n")
+end
+
+def models_table(models)
+  return '*No models found*' if models.none?
+
+  headers = ['Model', 'Provider', 'Context', 'Max Output', 'Standard Pricing (per 1M tokens)']
+  alignment = [':--', ':--', '--:', '--:', ':--']
+
+  rows = models.sort_by { |m| [m.provider, m.name] }.map do |model|
+    pricing = standard_pricing_display(model)
+
+    [
+      model.id,
+      model.provider,
+      model.context_window || '-',
+      model.max_output_tokens || '-',
+      pricing
+    ]
+  end
+
+  table = []
+  table << "| #{headers.join(' | ')} |"
+  table << "| #{alignment.join(' | ')} |"
+
+  rows.each do |row|
+    table << "| #{row.join(' | ')} |"
+  end
+
+  table.join("\n")
+end
+
+def standard_pricing_display(model)
+  pricing_data = model.pricing.to_h[:text_tokens]&.dig(:standard) || {}
+
+  if pricing_data.any?
+    parts = []
+
+    parts << "In: $#{format('%.2f', pricing_data[:input_per_million])}" if pricing_data[:input_per_million]
+
+    parts << "Out: $#{format('%.2f', pricing_data[:output_per_million])}" if pricing_data[:output_per_million]
+
+    if pricing_data[:cached_input_per_million]
+      parts << "Cache: $#{format('%.2f', pricing_data[:cached_input_per_million])}"
+    end
+
+    return parts.join(', ') if parts.any?
+  end
+
+  '-'
+end
+
+def generate_aliases # rubocop:disable Metrics/PerceivedComplexity
+  models = Hash.new { |h, k| h[k] = [] }
+
+  RubyLLM.models.all.each do |model|
+    models[model.provider] << model.id
+  end
+
+  aliases = {}
+
+  # OpenAI models
+  models['openai'].each do |model|
+    openrouter_model = "openai/#{model}"
+    next unless models['openrouter'].include?(openrouter_model)
+
+    alias_key = model.gsub('-latest', '')
+    aliases[alias_key] = {
+      'openai' => model,
+      'openrouter' => openrouter_model
+    }
+  end
+
+  anthropic_latest = group_anthropic_models_by_base_name(models['anthropic'])
+
+  anthropic_latest.each do |base_name, latest_model|
+    openrouter_variants = [
+      "anthropic/#{base_name}",
+      "anthropic/#{base_name.gsub(/-(\d)/, '.\1')}",
+      "anthropic/#{base_name.gsub(/claude-(\d+)-(\d+)/, 'claude-\1.\2')}",
+      "anthropic/#{base_name.gsub(/(\d+)-(\d+)/, '\1.\2')}"
+    ]
+
+    openrouter_model = openrouter_variants.find { |v| models['openrouter'].include?(v) }
+    bedrock_model = find_best_bedrock_model(latest_model, models['bedrock'])
+
+    next unless openrouter_model || bedrock_model || models['anthropic'].include?(latest_model)
+
+    aliases[base_name] = { 'anthropic' => latest_model }
+    aliases[base_name]['openrouter'] = openrouter_model if openrouter_model
+    aliases[base_name]['bedrock'] = bedrock_model if bedrock_model
+  end
+
+  models['bedrock'].each do |bedrock_model|
+    next unless bedrock_model.start_with?('anthropic.')
+    next unless bedrock_model =~ /anthropic\.(claude-[\d\.]+-[a-z]+)/
+
+    base_name = Regexp.last_match(1)
+    anthropic_name = base_name.tr('.', '-')
+
+    next if aliases[anthropic_name]
+
+    openrouter_variants = [
+      "anthropic/#{anthropic_name}",
+      "anthropic/#{base_name}"
+    ]
+
+    openrouter_model = openrouter_variants.find { |v| models['openrouter'].include?(v) }
+
+    aliases[anthropic_name] = { 'bedrock' => bedrock_model }
+    aliases[anthropic_name]['anthropic'] = anthropic_name if models['anthropic'].include?(anthropic_name)
+    aliases[anthropic_name]['openrouter'] = openrouter_model if openrouter_model
+  end
+
+  models['gemini'].each do |model|
+    openrouter_variants = [
+      "google/#{model}",
+      "google/#{model.gsub('gemini-', 'gemini-').tr('.', '-')}",
+      "google/#{model.gsub('gemini-', 'gemini-')}"
+    ]
+
+    openrouter_model = openrouter_variants.find { |v| models['openrouter'].include?(v) }
+    next unless openrouter_model
+
+    alias_key = model.gsub('-latest', '')
+    aliases[alias_key] = {
+      'gemini' => model,
+      'openrouter' => openrouter_model
+    }
+  end
+
+  models['deepseek'].each do |model|
+    openrouter_model = "deepseek/#{model}"
+    next unless models['openrouter'].include?(openrouter_model)
+
+    alias_key = model.gsub('-latest', '')
+    aliases[alias_key] = {
+      'deepseek' => model,
+      'openrouter' => openrouter_model
+    }
+  end
+
+  sorted_aliases = aliases.sort.to_h
+  File.write(RubyLLM::Aliases.aliases_file, JSON.pretty_generate(sorted_aliases))
+
+  puts "Generated #{sorted_aliases.size} aliases"
+end
+
+def group_anthropic_models_by_base_name(anthropic_models)
+  grouped = Hash.new { |h, k| h[k] = [] }
+
+  anthropic_models.each do |model|
+    base_name = extract_base_name(model)
+    grouped[base_name] << model
+  end
+
+  latest_models = {}
+  grouped.each do |base_name, model_list|
+    if model_list.size == 1
+      latest_models[base_name] = model_list.first
+    else
+      latest_model = model_list.max_by { |model| extract_date_from_model(model) }
+      latest_models[base_name] = latest_model
+    end
+  end
+
+  latest_models
+end
+
+def extract_base_name(model)
+  if model =~ /^(.+)-(\d{8})$/
+    Regexp.last_match(1)
+  else
+    model
+  end
+end
+
+def extract_date_from_model(model)
+  if model =~ /-(\d{8})$/
+    Regexp.last_match(1)
+  else
+    '00000000'
+  end
+end
+
+def find_best_bedrock_model(anthropic_model, bedrock_models) # rubocop:disable Metrics/PerceivedComplexity
+  base_pattern = case anthropic_model
+                 when 'claude-2.0', 'claude-2'
+                   'claude-v2'
+                 when 'claude-2.1'
+                   'claude-v2:1'
+                 when 'claude-instant-v1', 'claude-instant'
+                   'claude-instant'
+                 else
+                   extract_base_name(anthropic_model)
+                 end
+
+  matching_models = bedrock_models.select do |bedrock_model|
+    model_without_prefix = bedrock_model.sub(/^(?:us\.)?anthropic\./, '')
+    model_without_prefix.start_with?(base_pattern)
+  end
+
+  return nil if matching_models.empty?
+
+  begin
+    model_info = RubyLLM.models.find(anthropic_model)
+    target_context = model_info.context_window
+  rescue StandardError
+    target_context = nil
+  end
+
+  if target_context
+    target_k = target_context / 1000
+
+    with_context = matching_models.select do |m|
+      m.include?(":#{target_k}k") || m.include?(":0:#{target_k}k")
+    end
+
+    return with_context.first if with_context.any?
+  end
+
+  matching_models.min_by do |model|
+    context_priority = if model =~ /:(?:\d+:)?(\d+)k/
+                         -Regexp.last_match(1).to_i
+                       else
+                         0
+                       end
+
+    version_priority = if model =~ /-v(\d+):/
+                         -Regexp.last_match(1).to_i
+                       else
+                         0
+                       end
+
+    has_context_priority = model.include?('k') ? -1 : 0
+    [has_context_priority, context_priority, version_priority]
+  end
+end

data/lib/tasks/release.rake

@@ -1,6 +1,38 @@
 # frozen_string_literal: true
 
-namespace :release do
+namespace :release do # rubocop:disable Metrics/BlockLength
+  desc 'Prepare for release'
+  task :prepare do
+    Rake::Task['release:refresh_stale_cassettes'].invoke
+    sh 'overcommit --run'
+    Rake::Task['models'].invoke
+  end
+
+  desc 'Remove stale cassettes and re-record them'
+  task :refresh_stale_cassettes do
+    max_age_days = 1
+    cassette_dir = 'spec/fixtures/vcr_cassettes'
+
+    stale_count = 0
+    Dir.glob("#{cassette_dir}/**/*.yml").each do |cassette|
+      age_days = (Time.now - File.mtime(cassette)) / 86_400
+      next unless age_days > max_age_days
+
+      puts "Removing stale cassette: #{File.basename(cassette)} (#{age_days.round(1)} days old)"
+      File.delete(cassette)
+      stale_count += 1
+    end
+
+    if stale_count.positive?
+      puts "\n🗑️  Removed #{stale_count} stale cassettes"
+      puts '🔄 Re-recording cassettes...'
+      system('bundle exec rspec') || exit(1)
+      puts '✅ Cassettes refreshed!'
+    else
+      puts '✅ No stale cassettes found'
+    end
+  end
+
   desc 'Verify cassettes are fresh enough for release'
   task :verify_cassettes do
     max_age_days = 1
@@ -20,10 +52,13 @@ namespace :release do
 
     if stale_cassettes.any?
       puts "\n❌ Found stale cassettes (older than #{max_age_days} days):"
+      stale_files = []
       stale_cassettes.each do |c|
         puts "   - #{c[:file]} (#{c[:age]} days old)"
+        stale_files << File.join(cassette_dir, '**', c[:file])
       end
-      puts "\nRun locally: bundle exec rspec"
+
+      puts "\nRun locally: bundle exec rake release:refresh_stale_cassettes"
       exit 1
     else
       puts "✅ All cassettes are fresh (< #{max_age_days} days old)"

data/lib/tasks/vcr.rake

@@ -2,9 +2,7 @@
 
 require 'dotenv/load'
 
-# Helper functions at the top level
 def record_all_cassettes(cassette_dir)
-  # Re-record all cassettes
   FileUtils.rm_rf(cassette_dir)
   FileUtils.mkdir_p(cassette_dir)
 
@@ -14,10 +12,8 @@ def record_all_cassettes(cassette_dir)
 end
 
 def record_for_providers(providers, cassette_dir)
-  # Get the list of available providers from RubyLLM itself
   all_providers = RubyLLM::Provider.providers.keys.map(&:to_s)
 
-  # Check for valid providers
   if providers.empty?
     puts "Please specify providers or 'all'. Example: rake vcr:record[openai,anthropic]"
     puts "Available providers: #{all_providers.join(', ')}"
@@ -31,7 +27,6 @@ def record_for_providers(providers, cassette_dir)
     return
   end
 
-  # Find and delete matching cassettes
   cassettes_to_delete = find_matching_cassettes(cassette_dir, providers)
 
   if cassettes_to_delete.empty?
@@ -54,9 +49,7 @@ def find_matching_cassettes(dir, providers)
   Dir.glob("#{dir}/**/*.yml").each do |file|
     basename = File.basename(file)
 
-    # Precise matching to avoid cross-provider confusion
     providers.each do |provider|
-      # Match only exact provider prefixes
       next unless basename =~ /^[^_]*_#{provider}_/ || # For first section like "chat_openai_"
                   basename =~ /_#{provider}_[^_]+_/    # For middle sections like "_openai_gpt4_"
 

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: ruby_llm
 version: !ruby/object:Gem::Version
-  version: 1.6.2
+  version: 1.6.4
 platform: ruby
 authors:
 - Carmine Paolino
@@ -236,9 +236,7 @@ files:
 - lib/ruby_llm/tool_call.rb
 - lib/ruby_llm/utils.rb
 - lib/ruby_llm/version.rb
-- lib/tasks/aliases.rake
-- lib/tasks/models_docs.rake
-- lib/tasks/models_update.rake
+- lib/tasks/models.rake
 - lib/tasks/release.rake
 - lib/tasks/vcr.rake
 homepage: https://rubyllm.com