ruby_llm 0.1.0.pre30 → 0.1.0.pre31

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 33b602ae544211e79207cf51bbfad5f1ed9c78cd1d9710d4f57916c95377bb60
-  data.tar.gz: 3fac0d6aed252218d7f7bd8e460f7b281c0306ca066e0194193e11a44ac8400d
+  metadata.gz: a3c91f92537c2e154f458422c8447345cf1d230da68fb982df3b45b102fe985d
+  data.tar.gz: fd24e6187a456c0ec7ade007cf39b0766e41da58bb8184a0567939ea1c048e75
 SHA512:
-  metadata.gz: fa21a4d89b3704f384c257a7bcca71e42b0442b2f660ddd2c93d108f77bcf547b5d1c05101d85269abc5aab8fbe37e3bec3da8a4fd09d77f8184ed923fea0986
-  data.tar.gz: 5295d0c1ec5660e4ce9e7388836bbc37d82eae0ccce77a4f601f59ed91933cb2dc31b84f79f5aab3271d5b8b318aa190ca0224926a28c69a0e409ed9035db4c8
+  metadata.gz: 6cba0bb735838fb5700d504e8665842815001bbab5c8a1a829131eaefef6e10b7633fde20be63bf4174eaac045298590489a777572a5bbea9141a32392bc7a36
+  data.tar.gz: a2707917f85c6fdd31c8f61f6ae9d3414ac385d17bf017335572cf61be03bd6dcd344fd277e207fcbc0954e3cc445cbdeb40b17d432de0800191a31b0dab7383

data/.github/workflows/{gem-push.yml → cicd.yml}

@@ -1,18 +1,46 @@
-name: Ruby Gem
+name: CI
 
 on:
   push:
     branches: [ "main" ]
   pull_request:
     branches: [ "main" ]
+  workflow_call:
 
 jobs:
   test:
-    uses: ./.github/workflows/test.yml
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        ruby-version: ['3.1']
+
+    steps:
+    - uses: actions/checkout@v4
+
+    - name: Set up Ruby
+      uses: ruby/setup-ruby@v1
+      with:
+        ruby-version: ${{ matrix.ruby-version }}
+        bundler-cache: true
+
+    - name: Install dependencies
+      run: bundle install
+
+    - name: Check code format
+      run: bundle exec rubocop
+
+    - name: Run tests
+      env:
+        OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+        ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
+        GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
+        DEEPSEEK_API_KEY: ${{ secrets.DEEPSEEK_API_KEY }}
+      run: bundle exec rspec
 
-  build:
-    needs: test # This ensures tests must pass before building/publishing
+  publish:
     name: Build + Publish
+    needs: test
+    if: github.event_name == 'push' && github.ref == 'refs/heads/main'
     runs-on: ubuntu-latest
     permissions:
       contents: read

data/.rspec_status

@@ -0,0 +1,27 @@
+example_id                                     | status | run_time        |
+---------------------------------------------- | ------ | --------------- |
+./spec/integration/chat_spec.rb[1:1:1:1]       | passed | 0.5826 seconds  |
+./spec/integration/chat_spec.rb[1:1:1:2]       | passed | 3.43 seconds    |
+./spec/integration/chat_spec.rb[1:1:2:1]       | passed | 0.49383 seconds |
+./spec/integration/chat_spec.rb[1:1:2:2]       | passed | 1.36 seconds    |
+./spec/integration/chat_spec.rb[1:1:3:1]       | passed | 0.74049 seconds |
+./spec/integration/chat_spec.rb[1:1:3:2]       | passed | 4.11 seconds    |
+./spec/integration/embeddings_spec.rb[1:1:1:1] | passed | 0.2892 seconds  |
+./spec/integration/embeddings_spec.rb[1:1:1:2] | passed | 0.31644 seconds |
+./spec/integration/embeddings_spec.rb[1:1:2:1] | passed | 0.89277 seconds |
+./spec/integration/embeddings_spec.rb[1:1:2:2] | passed | 1.55 seconds    |
+./spec/integration/error_handling_spec.rb[1:1] | passed | 0.21297 seconds |
+./spec/integration/rails_spec.rb[1:1]          | passed | 4.05 seconds    |
+./spec/integration/rails_spec.rb[1:2]          | passed | 1.82 seconds    |
+./spec/integration/streaming_spec.rb[1:1:1:1]  | passed | 0.58445 seconds |
+./spec/integration/streaming_spec.rb[1:1:1:2]  | passed | 6.04 seconds    |
+./spec/integration/streaming_spec.rb[1:1:2:1]  | passed | 0.47171 seconds |
+./spec/integration/streaming_spec.rb[1:1:2:2]  | passed | 2.39 seconds    |
+./spec/integration/streaming_spec.rb[1:1:3:1]  | passed | 0.72016 seconds |
+./spec/integration/streaming_spec.rb[1:1:3:2]  | passed | 3.59 seconds    |
+./spec/integration/tools_spec.rb[1:1:1:1]      | passed | 3.1 seconds     |
+./spec/integration/tools_spec.rb[1:1:1:2]      | passed | 7.04 seconds    |
+./spec/integration/tools_spec.rb[1:1:2:1]      | passed | 1.42 seconds    |
+./spec/integration/tools_spec.rb[1:1:2:2]      | passed | 2.24 seconds    |
+./spec/integration/tools_spec.rb[1:1:3:1]      | passed | 2.16 seconds    |
+./spec/integration/tools_spec.rb[1:1:3:2]      | passed | 5.26 seconds    |

data/lib/ruby_llm/active_record/acts_as.rb

@@ -73,22 +73,22 @@ module RubyLLM
       end
 
       def to_llm
-        chat = RubyLLM.chat(model: model_id)
+        @chat ||= RubyLLM.chat(model: model_id)
 
         # Load existing messages into chat
         messages.each do |msg|
-          chat.add_message(msg.to_llm)
+          @chat.add_message(msg.to_llm)
         end
 
         # Set up message persistence
-        chat.on_new_message { persist_new_message }
-            .on_end_message { |msg| persist_message_completion(msg) }
+        @chat.on_new_message { persist_new_message }
+             .on_end_message { |msg| persist_message_completion(msg) }
       end
 
       def ask(message, &block)
         message = { role: :user, content: message }
         messages.create!(**message)
-        chat.complete(&block)
+        to_llm.complete(&block)
       end
 
       alias say ask

data/lib/ruby_llm/chat.rb

@@ -25,8 +25,8 @@ module RubyLLM
       }
     end
 
-    def ask(message, &block)
-      add_message role: :user, content: message
+    def ask(message = nil, with: {}, &block)
+      add_message role: :user, content: Content.new(message, with)
       complete(&block)
     end
 

data/lib/ruby_llm/configuration.rb

@@ -16,10 +16,12 @@ module RubyLLM
                   :deepseek_api_key,
                   :default_model,
                   :default_embedding_model,
-                  :request_timeout
+                  :request_timeout,
+                  :max_retries
 
     def initialize
       @request_timeout = 120
+      @max_retries = 3
       @default_model = 'gpt-4o-mini'
       @default_embedding_model = 'text-embedding-3-small'
     end

data/lib/ruby_llm/content.rb

@@ -0,0 +1,79 @@
+# frozen_string_literal: true
+
+module RubyLLM
+  # Represents the content received from the LLM
+  class Content
+    def initialize(text = nil, attachments = {})
+      @parts = []
+      @parts << { type: 'text', text: text } unless text.nil? || text.empty?
+
+      Array(attachments[:image]).each do |source|
+        @parts << attach_image(source)
+      end
+
+      Array(attachments[:audio]).each do |source|
+        @parts << attach_audio(source)
+      end
+    end
+
+    def to_a
+      return if @parts.empty?
+
+      @parts
+    end
+
+    def format
+      return @parts.first[:text] if @parts.size == 1 && @parts.first[:type] == 'text'
+
+      to_a
+    end
+
+    private
+
+    def attach_image(source) # rubocop:disable Metrics/MethodLength
+      source = File.expand_path(source) unless source.start_with?('http')
+
+      return { type: 'image_url', image_url: { url: source } } if source.start_with?('http')
+
+      data = Base64.strict_encode64(File.read(source))
+      mime_type = mime_type_for(source)
+
+      {
+        type: 'image',
+        source: {
+          type: 'base64',
+          media_type: mime_type,
+          data: data
+        }
+      }
+    end
+
+    def attach_audio(source)
+      source = File.expand_path(source) unless source.start_with?('http')
+      data = encode_file(source)
+      format = File.extname(source).delete('.') || 'wav'
+
+      {
+        type: 'input_audio',
+        input_audio: {
+          data: data,
+          format: format
+        }
+      }
+    end
+
+    def encode_file(source)
+      if source.start_with?('http')
+        response = Faraday.get(source)
+        Base64.strict_encode64(response.body)
+      else
+        Base64.strict_encode64(File.read(source))
+      end
+    end
+
+    def mime_type_for(path)
+      ext = File.extname(path).delete('.')
+      "image/#{ext}"
+    end
+  end
+end

data/lib/ruby_llm/embedding.rb

@@ -3,10 +3,16 @@
 module RubyLLM
   # Core embedding interface. Provides a clean way to generate embeddings
   # from text using various provider models.
-  module Embedding
-    module_function
+  class Embedding
+    attr_reader :vectors, :model, :input_tokens
 
-    def embed(text, model: nil)
+    def initialize(vectors:, model:, input_tokens: 0)
+      @vectors = vectors
+      @model = model
+      @input_tokens = input_tokens
+    end
+
+    def self.embed(text, model: nil)
       model_id = model || RubyLLM.config.default_embedding_model
       Models.find(model_id)
 

data/lib/ruby_llm/message.rb

@@ -11,7 +11,7 @@ module RubyLLM
 
     def initialize(options = {})
       @role = options[:role].to_sym
-      @content = options[:content]
+      @content = normalize_content(options[:content])
       @tool_calls = options[:tool_calls]
       @input_tokens = options[:input_tokens]
       @output_tokens = options[:output_tokens]
@@ -47,6 +47,14 @@ module RubyLLM
 
     private
 
+    def normalize_content(content)
+      case content
+      when Content then content.format
+      when String then Content.new(content).format
+      else content
+      end
+    end
+
     def ensure_valid_role
       raise InvalidRoleError, "Expected role to be one of: #{ROLES.join(', ')}" unless ROLES.include?(role)
     end

data/lib/ruby_llm/models.json

@@ -727,7 +727,7 @@
     "family": "gemini20_flash",
     "supports_vision": true,
     "supports_functions": true,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.1,
     "output_price_per_million": 0.4,
     "metadata": {
@@ -746,7 +746,7 @@
     "family": "gemini20_flash",
     "supports_vision": true,
     "supports_functions": true,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.1,
     "output_price_per_million": 0.4,
     "metadata": {
@@ -765,7 +765,7 @@
     "family": "gemini20_flash",
     "supports_vision": true,
     "supports_functions": true,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.1,
     "output_price_per_million": 0.4,
     "metadata": {
@@ -784,7 +784,7 @@
     "family": "gemini20_flash",
     "supports_vision": true,
     "supports_functions": true,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.1,
     "output_price_per_million": 0.4,
     "metadata": {
@@ -803,7 +803,7 @@
     "family": "gemini20_flash",
     "supports_vision": true,
     "supports_functions": true,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.1,
     "output_price_per_million": 0.4,
     "metadata": {
@@ -822,7 +822,7 @@
     "family": "gemini20_flash",
     "supports_vision": true,
     "supports_functions": true,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.1,
     "output_price_per_million": 0.4,
     "metadata": {
@@ -841,7 +841,7 @@
     "family": "gemini20_flash_lite",
     "supports_vision": true,
     "supports_functions": false,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.075,
     "output_price_per_million": 0.3,
     "metadata": {
@@ -860,7 +860,7 @@
     "family": "gemini20_flash_lite",
     "supports_vision": true,
     "supports_functions": false,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.075,
     "output_price_per_million": 0.3,
     "metadata": {
@@ -879,7 +879,7 @@
     "family": "gemini20_flash",
     "supports_vision": true,
     "supports_functions": true,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.1,
     "output_price_per_million": 0.4,
     "metadata": {
@@ -898,7 +898,7 @@
     "family": "gemini20_flash",
     "supports_vision": true,
     "supports_functions": true,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.1,
     "output_price_per_million": 0.4,
     "metadata": {
@@ -917,7 +917,7 @@
     "family": "gemini20_flash",
     "supports_vision": true,
     "supports_functions": true,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.1,
     "output_price_per_million": 0.4,
     "metadata": {
@@ -936,7 +936,7 @@
     "family": "gemini20_flash",
     "supports_vision": true,
     "supports_functions": true,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.1,
     "output_price_per_million": 0.4,
     "metadata": {
@@ -955,7 +955,7 @@
     "family": "other",
     "supports_vision": true,
     "supports_functions": true,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.075,
     "output_price_per_million": 0.3,
     "metadata": {
@@ -974,7 +974,7 @@
     "family": "other",
     "supports_vision": true,
     "supports_functions": true,
-    "supports_json_mode": false,
+    "supports_json_mode": true,
     "input_price_per_million": 0.075,
     "output_price_per_million": 0.3,
     "metadata": {

data/lib/ruby_llm/provider.rb

@@ -5,15 +5,11 @@ module RubyLLM
   # Handles the complexities of API communication, streaming responses,
   # and error handling so individual providers can focus on their unique features.
   module Provider
-    def self.included(base)
-      base.include(InstanceMethods)
-    end
-
     # Common functionality for all LLM providers. Implements the core provider
     # interface so specific providers only need to implement a few key methods.
-    module InstanceMethods
+    module Methods
       def complete(messages, tools:, temperature:, model:, &block)
-        payload = build_payload messages, tools: tools, temperature: temperature, model: model, stream: block_given?
+        payload = render_payload messages, tools: tools, temperature: temperature, model: model, stream: block_given?
 
         if block_given?
           stream_response payload, &block
@@ -31,7 +27,7 @@ module RubyLLM
       end
 
       def embed(text, model:)
-        payload = build_embedding_payload text, model: model
+        payload = render_embedding_payload text, model: model
         response = post embedding_url, payload
         parse_embedding_response response
       end
@@ -63,9 +59,29 @@ module RubyLLM
         end
       end
 
-      def connection
+      def connection # rubocop:disable Metrics/MethodLength
         @connection ||= Faraday.new(api_base) do |f|
           f.options.timeout = RubyLLM.config.request_timeout
+
+          # Add retry middleware before request/response handling
+          f.request :retry, {
+            max: RubyLLM.config.max_retries,
+            interval: 0.05,
+            interval_randomness: 0.5,
+            backoff_factor: 2,
+            exceptions: [
+              Errno::ETIMEDOUT,
+              Timeout::Error,
+              Faraday::TimeoutError,
+              Faraday::ConnectionFailed,
+              Faraday::RetriableResponse,
+              RubyLLM::RateLimitError,
+              RubyLLM::ServerError,
+              RubyLLM::ServiceUnavailableError
+            ],
+            retry_statuses: [429, 500, 502, 503, 504]
+          }
+
           f.request :json
           f.response :json
           f.adapter Faraday.default_adapter
@@ -111,9 +127,16 @@ module RubyLLM
       maybe_json
     end
 
+    def parse_error(response)
+      return if response.body.empty?
+
+      body = try_parse_json(response.body)
+      body.is_a?(Hash) ? body.dig('error', 'message') : body
+    end
+
     def capabilities
       provider_name = self.class.name.split('::').last
-      RubyLLM.const_get "ModelCapabilities::#{provider_name}"
+      provider_name::Capabilities
     end
 
     def slug
@@ -121,15 +144,17 @@ module RubyLLM
     end
 
     class << self
-      def register(name, provider_class)
-        providers[name.to_sym] = provider_class
+      def extended(base)
+        base.extend(Methods)
+      end
+
+      def register(name, provider_module)
+        providers[name.to_sym] = provider_module
       end
 
       def for(model)
         model_info = Models.find(model)
-        provider_class = providers[model_info.provider.to_sym]
-
-        provider_class.new
+        providers[model_info.provider.to_sym]
       end
 
       def providers

data/lib/ruby_llm/providers/anthropic/capabilities.rb

@@ -0,0 +1,81 @@
+# frozen_string_literal: true
+
+module RubyLLM
+  module Providers
+    module Anthropic
+      # Determines capabilities and pricing for Anthropic models
+      module Capabilities
+        module_function
+
+        def determine_context_window(model_id)
+          case model_id
+          when /claude-3/ then 200_000
+          else 100_000
+          end
+        end
+
+        def determine_max_tokens(model_id)
+          case model_id
+          when /claude-3-5/ then 8_192
+          else 4_096
+          end
+        end
+
+        def get_input_price(model_id)
+          PRICES.dig(model_family(model_id), :input) || default_input_price
+        end
+
+        def get_output_price(model_id)
+          PRICES.dig(model_family(model_id), :output) || default_output_price
+        end
+
+        def supports_vision?(model_id)
+          return false if model_id.match?(/claude-3-5-haiku/)
+          return false if model_id.match?(/claude-[12]/)
+
+          true
+        end
+
+        def supports_functions?(model_id)
+          model_id.include?('claude-3')
+        end
+
+        def supports_json_mode?(model_id)
+          model_id.include?('claude-3')
+        end
+
+        def model_family(model_id)
+          case model_id
+          when /claude-3-5-sonnet/  then :claude35_sonnet
+          when /claude-3-5-haiku/   then :claude35_haiku
+          when /claude-3-opus/      then :claude3_opus
+          when /claude-3-sonnet/    then :claude3_sonnet
+          when /claude-3-haiku/     then :claude3_haiku
+          else :claude2
+          end
+        end
+
+        def model_type(_)
+          'chat'
+        end
+
+        PRICES = {
+          claude35_sonnet: { input: 3.0, output: 15.0 }, # $3.00/$15.00 per million tokens
+          claude35_haiku: { input: 0.80, output: 4.0 }, # $0.80/$4.00 per million tokens
+          claude3_opus: { input: 15.0, output: 75.0 }, # $15.00/$75.00 per million tokens
+          claude3_sonnet: { input: 3.0, output: 15.0 }, # $3.00/$15.00 per million tokens
+          claude3_haiku: { input: 0.25, output: 1.25 }, # $0.25/$1.25 per million tokens
+          claude2: { input: 3.0,  output: 15.0 } # Default pricing for Claude 2.x models
+        }.freeze
+
+        def default_input_price
+          3.0
+        end
+
+        def default_output_price
+          15.0
+        end
+      end
+    end
+  end
+end

data/lib/ruby_llm/providers/anthropic/chat.rb

@@ -0,0 +1,86 @@
+# frozen_string_literal: true
+
+module RubyLLM
+  module Providers
+    module Anthropic
+      # Chat methods of the OpenAI API integration
+      module Chat
+        private
+
+        def completion_url
+          '/v1/messages'
+        end
+
+        def render_payload(messages, tools:, temperature:, model:, stream: false)
+          {
+            model: model,
+            messages: messages.map { |msg| format_message(msg) },
+            temperature: temperature,
+            stream: stream,
+            max_tokens: RubyLLM.models.find(model).max_tokens
+          }.tap do |payload|
+            payload[:tools] = tools.values.map { |t| function_for(t) } if tools.any?
+          end
+        end
+
+        def parse_completion_response(response)
+          data = response.body
+          content_blocks = data['content'] || []
+
+          text_content = extract_text_content(content_blocks)
+          tool_use = find_tool_use(content_blocks)
+
+          build_message(data, text_content, tool_use)
+        end
+
+        def extract_text_content(blocks)
+          text_blocks = blocks.select { |c| c['type'] == 'text' }
+          text_blocks.map { |c| c['text'] }.join('')
+        end
+
+        def build_message(data, content, tool_use)
+          Message.new(
+            role: :assistant,
+            content: content,
+            tool_calls: parse_tool_calls(tool_use),
+            input_tokens: data.dig('usage', 'input_tokens'),
+            output_tokens: data.dig('usage', 'output_tokens'),
+            model_id: data['model']
+          )
+        end
+
+        def format_message(msg)
+          if msg.tool_call?
+            format_tool_call(msg)
+          elsif msg.tool_result?
+            format_tool_result(msg)
+          else
+            format_basic_message(msg)
+          end
+        end
+
+        def format_basic_message(msg)
+          {
+            role: convert_role(msg.role),
+            content: msg.content
+          }
+        end
+
+        def convert_role(role)
+          case role
+          when :tool then 'user'
+          when :user then 'user'
+          else 'assistant'
+          end
+        end
+
+        def format_text_block(content)
+          {
+            type: 'text',
+            text: content
+          }
+        end
+      end
+    end
+  end
+end

data/lib/ruby_llm/providers/anthropic/embeddings.rb

@@ -0,0 +1,20 @@
+# frozen_string_literal: true
+
+module RubyLLM
+  module Providers
+    module Anthropic
+      # Embeddings methods of the Anthropic API integration
+      module Embeddings
+        private
+
+        def embed
+          raise Error "Anthropic doesn't support embeddings"
+        end
+
+        alias render_embedding_payload embed
+        alias embedding_url embed
+        alias parse_embedding_response embed
+      end
+    end
+  end
+end