durable-llm 0.1.4 → 0.1.5

data/lib/durable/llm/providers/azure_openai.rb

@@ -0,0 +1,347 @@
+# frozen_string_literal: true
+
+# Azure OpenAI provider implementation for Durable LLM
+
+require 'faraday'
+require 'json'
+require 'durable/llm/errors'
+require 'durable/llm/providers/base'
+require 'event_stream_parser'
+
+module Durable
+  module Llm
+    module Providers
+      # Azure OpenAI provider for accessing Azure OpenAI's language models
+      #
+      # This provider implements the Azure OpenAI API for chat completions,
+      # embeddings, and streaming. It handles authentication via API keys,
+      # deployment-based routing, and response normalization.
+      class AzureOpenai < Durable::Llm::Providers::Base
+        BASE_URL_TEMPLATE = 'https://%s.openai.azure.com/openai/deployments/%s'
+
+        def default_api_key
+          begin
+            Durable::Llm.configuration.azure_openai&.api_key
+          rescue NoMethodError
+            nil
+          end || ENV['AZURE_OPENAI_API_KEY']
+        end
+
+        attr_accessor :api_key, :resource_name, :api_version
+
+        def initialize(api_key: nil, resource_name: nil, api_version: '2024-02-01')
+          super(api_key: api_key)
+          @resource_name = resource_name || ENV['AZURE_OPENAI_RESOURCE_NAME']
+          @api_version = api_version
+          # NOTE: BASE_URL will be constructed per request since deployment is in model
+        end
+
+        def completion(options)
+          model = options.delete(:model) || options.delete('model')
+          base_url = format(BASE_URL_TEMPLATE, @resource_name, model)
+          conn = build_connection(base_url)
+
+          response = conn.post('chat/completions') do |req|
+            req.headers['api-key'] = @api_key
+            req.params['api-version'] = @api_version
+            req.body = options
+          end
+
+          handle_response(response)
+        end
+
+        def embedding(model:, input:, **options)
+          base_url = format(BASE_URL_TEMPLATE, @resource_name, model)
+          conn = build_connection(base_url)
+
+          response = conn.post('embeddings') do |req|
+            req.headers['api-key'] = @api_key
+            req.params['api-version'] = @api_version
+            req.body = { input: input, **options }
+          end
+
+          handle_response(response, AzureOpenaiEmbeddingResponse)
+        end
+
+        def models
+          # Azure OpenAI doesn't have a public models endpoint, return hardcoded list
+          [
+            # GPT-5 series
+            'gpt-5',
+            'gpt-5-mini',
+            'gpt-5-nano',
+            'gpt-5-chat',
+            'gpt-5-codex',
+            'gpt-5-pro',
+            # GPT-4.1 series
+            'gpt-4.1',
+            'gpt-4.1-mini',
+            'gpt-4.1-nano',
+            # GPT-4o series
+            'gpt-4o',
+            'gpt-4o-mini',
+            'gpt-4o-audio-preview',
+            'gpt-4o-mini-audio-preview',
+            'gpt-4o-realtime-preview',
+            'gpt-4o-mini-realtime-preview',
+            'gpt-4o-transcribe',
+            'gpt-4o-mini-transcribe',
+            'gpt-4o-mini-tts',
+            # GPT-4 Turbo
+            'gpt-4-turbo',
+            # GPT-4
+            'gpt-4',
+            'gpt-4-32k',
+            # GPT-3.5
+            'gpt-3.5-turbo',
+            'gpt-35-turbo',
+            'gpt-35-turbo-instruct',
+            # O-series
+            'o3',
+            'o3-mini',
+            'o3-pro',
+            'o4-mini',
+            'o1',
+            'o1-mini',
+            'o1-preview',
+            'codex-mini',
+            # Embeddings
+            'text-embedding-ada-002',
+            'text-embedding-3-small',
+            'text-embedding-3-large',
+            # Audio
+            'whisper',
+            'gpt-4o-transcribe',
+            'gpt-4o-mini-transcribe',
+            'tts',
+            'tts-hd',
+            'gpt-4o-mini-tts',
+            # Image generation
+            'dall-e-3',
+            'gpt-image-1',
+            'gpt-image-1-mini',
+            # Video generation
+            'sora',
+            # Other
+            'model-router',
+            'computer-use-preview',
+            'gpt-oss-120b',
+            'gpt-oss-20b'
+          ]
+        end
+
+        def self.stream?
+          true
+        end
+
+        def stream(options)
+          model = options[:model] || options['model']
+          base_url = format(BASE_URL_TEMPLATE, @resource_name, model)
+          conn = build_connection(base_url)
+
+          options[:stream] = true
+          options['temperature'] = options['temperature'].to_f if options['temperature']
+
+          response = conn.post('chat/completions') do |req|
+            setup_stream_request(req, options) do |chunk|
+              yield AzureOpenaiStreamResponse.new(chunk)
+            end
+          end
+
+          handle_response(response)
+        end
+
+        def setup_stream_request(req, options)
+          req.headers['api-key'] = @api_key
+          req.params['api-version'] = @api_version
+          req.headers['Accept'] = 'text/event-stream'
+          req.body = options
+
+          user_proc = proc do |chunk, _size, _total|
+            yield chunk
+          end
+
+          req.options.on_data = to_json_stream(user_proc: user_proc)
+        end
+
+        private
+
+        def build_connection(base_url)
+          Faraday.new(url: base_url) do |faraday|
+            faraday.request :json
+            faraday.response :json
+            faraday.adapter Faraday.default_adapter
+          end
+        end
+
+        # CODE-FROM: ruby-openai @ https://github.com/alexrudall/ruby-openai/blob/main/lib/openai/http.rb
+        # MIT License: https://github.com/alexrudall/ruby-openai/blob/main/LICENSE.md
+        def to_json_stream(user_proc:)
+          parser = EventStreamParser::Parser.new
+
+          proc do |chunk, _bytes, env|
+            if env && env.status != 200
+              raise_error = Faraday::Response::RaiseError.new
+              raise_error.on_complete(env.merge(body: try_parse_json(chunk)))
+            end
+
+            parser.feed(chunk) do |_type, data|
+              user_proc.call(JSON.parse(data)) unless data == '[DONE]'
+            end
+          end
+        end
+
+        def try_parse_json(maybe_json)
+          JSON.parse(maybe_json)
+        rescue JSON::ParserError
+          maybe_json
+        end
+
+        # END-CODE-FROM
+
+        def handle_response(response, response_class = AzureOpenaiResponse)
+          case response.status
+          when 200..299
+            response_class.new(response.body)
+          else
+            raise_error(response)
+          end
+        end
+
+        def raise_error(response)
+          error_class = case response.status
+                        when 401 then Durable::Llm::AuthenticationError
+                        when 429 then Durable::Llm::RateLimitError
+                        when 400..499 then Durable::Llm::InvalidRequestError
+                        when 500..599 then Durable::Llm::ServerError
+                        else Durable::Llm::APIError
+                        end
+
+          message = if error_class == Durable::Llm::APIError
+                      "Unexpected response code: #{response.status}"
+                    else
+                      parse_error_message(response)
+                    end
+
+          raise error_class, message
+        end
+
+        def parse_error_message(response)
+          body = begin
+            JSON.parse(response.body)
+          rescue StandardError
+            nil
+          end
+          message = body&.dig('error', 'message') || response.body
+          "#{response.status} Error: #{message}"
+        end
+
+        # Response wrapper for Azure OpenAI completion API responses
+        class AzureOpenaiResponse
+          attr_reader :raw_response
+
+          def initialize(response)
+            @raw_response = response
+          end
+
+          def choices
+            @raw_response['choices'].map { |choice| AzureOpenaiChoice.new(choice) }
+          end
+
+          def data
+            @raw_response['data']
+          end
+
+          def to_s
+            choices.map(&:to_s).join(' ')
+          end
+        end
+
+        # Choice wrapper for Azure OpenAI API responses
+        class AzureOpenaiChoice
+          attr_reader :message, :finish_reason
+
+          def initialize(choice)
+            @message = AzureOpenaiMessage.new(choice['message'])
+            @finish_reason = choice['finish_reason']
+          end
+
+          def to_s
+            @message.to_s
+          end
+        end
+
+        # Message wrapper for Azure OpenAI API responses
+        class AzureOpenaiMessage
+          attr_reader :role, :content
+
+          def initialize(message)
+            @role = message['role']
+            @content = message['content']
+          end
+
+          def to_s
+            @content
+          end
+        end
+
+        # Stream response wrapper for Azure OpenAI streaming API
+        class AzureOpenaiStreamResponse
+          attr_reader :choices
+
+          def initialize(parsed)
+            @choices = AzureOpenaiStreamChoice.new(parsed['choices'])
+          end
+
+          def to_s
+            @choices.to_s
+          end
+        end
+
+        # Embedding response wrapper for Azure OpenAI embedding API
+        class AzureOpenaiEmbeddingResponse
+          attr_reader :embedding
+
+          def initialize(data)
+            @embedding = data.dig('data', 0, 'embedding')
+          end
+
+          def to_a
+            @embedding
+          end
+        end
+
+        # Stream choice wrapper for Azure OpenAI streaming responses
+        class AzureOpenaiStreamChoice
+          attr_reader :delta, :finish_reason
+
+          def initialize(choice)
+            @choice = [choice].flatten.first
+            @delta = AzureOpenaiStreamDelta.new(@choice['delta'])
+            @finish_reason = @choice['finish_reason']
+          end
+
+          def to_s
+            @delta.to_s
+          end
+        end
+
+        # Stream delta wrapper for Azure OpenAI streaming responses
+        class AzureOpenaiStreamDelta
+          attr_reader :role, :content
+
+          def initialize(delta)
+            @role = delta['role']
+            @content = delta['content']
+          end
+
+          def to_s
+            @content || ''
+          end
+        end
+      end
+    end
+  end
+end
+
+# Copyright (c) 2025 Durable Programming, LLC. All rights reserved.

data/lib/durable/llm/providers/base.rb

@@ -1,21 +1,65 @@
+# frozen_string_literal: true
+
+require 'json'
+require 'fileutils'
+
+# This file defines the abstract base class for all LLM providers in the Durable gem,
+# establishing a common interface and shared functionality that all provider implementations
+# must follow. It defines required methods like completion, models, and streaming capabilities,
+# provides caching mechanisms for model lists, handles default API key resolution, and includes
+# stub implementations for optional features like embeddings. The base class ensures consistency
+# across different LLM providers while allowing each provider to implement their specific API
+# communication patterns and response handling.
+
 module Durable
   module Llm
     module Providers
+      # Abstract base class for all LLM providers
+      #
+      # This class defines the common interface that all LLM provider implementations must follow.
+      # It provides default implementations for caching model lists, handling API keys, and stub
+      # implementations for optional features.
+      #
+      # Subclasses must implement the following methods:
+      # - default_api_key
+      # - completion
+      # - models
+      # - handle_response
+      #
+      # Subclasses may override:
+      # - stream?
+      # - stream
+      # - embedding
       class Base
+        # @return [String, nil] The default API key for this provider, or nil if not configured
+        # @raise [NotImplementedError] Subclasses must implement this method
         def default_api_key
           raise NotImplementedError, 'Subclasses must implement default_api_key'
         end
 
+        # @!attribute [rw] api_key
+        #   @return [String, nil] The API key used for authentication
         attr_accessor :api_key
 
+        # Initializes a new provider instance
+        #
+        # @param api_key [String, nil] The API key to use for authentication. If nil, uses default_api_key
         def initialize(api_key: nil)
           @api_key = api_key || default_api_key
         end
 
+        # Performs a completion request
+        #
+        # @param options [Hash] The completion options including model, messages, etc.
+        # @return [Object] The completion response object
+        # @raise [NotImplementedError] Subclasses must implement this method
         def completion(options)
           raise NotImplementedError, 'Subclasses must implement completion'
         end
 
+        # Retrieves the list of available models, with caching
+        #
+        # @return [Array<String>] The list of available model names
         def self.models
           cache_dir = File.expand_path("#{Dir.home}/.local/durable-llm/cache")
 
@@ -29,33 +73,69 @@ module Durable
             JSON.parse(File.read(cache_file))
           else
             models = new.models
-            File.write(cache_file, JSON.generate(models)) if models.length > 0
+            File.write(cache_file, JSON.generate(models)) if models.length.positive?
             models
           end
         end
 
+        # Returns the list of supported option names for completions
+        #
+        # @return [Array<String>] The supported option names
+        def self.options
+          %w[temperature max_tokens top_p frequency_penalty presence_penalty]
+        end
+
+        # Retrieves the list of available models for this provider instance
+        #
+        # @return [Array<String>] The list of available model names
+        # @raise [NotImplementedError] Subclasses must implement this method
         def models
           raise NotImplementedError, 'Subclasses must implement models'
         end
 
+        # Checks if this provider class supports streaming
+        #
+        # @return [Boolean] True if streaming is supported, false otherwise
         def self.stream?
           false
         end
 
+        # Checks if this provider instance supports streaming
+        #
+        # @return [Boolean] True if streaming is supported, false otherwise
         def stream?
           self.class.stream?
         end
 
+        # Performs a streaming completion request
+        #
+        # @param options [Hash] The stream options including model, messages, etc.
+        # @yield [Object] Yields stream response chunks as they arrive
+        # @return [Object] The final response object
+        # @raise [NotImplementedError] Subclasses must implement this method
         def stream(options, &block)
           raise NotImplementedError, 'Subclasses must implement stream'
         end
 
+        # Performs an embedding request
+        #
+        # @param model [String] The model to use for generating embeddings
+        # @param input [String, Array<String>] The input text(s) to embed
+        # @param options [Hash] Additional options for the embedding request
+        # @return [Object] The embedding response object
+        # @raise [NotImplementedError] Subclasses must implement this method
         def embedding(model:, input:, **options)
           raise NotImplementedError, 'Subclasses must implement embedding'
         end
 
         private
 
+        # Handles the raw response from the API, processing errors and returning normalized response
+        #
+        # @param response [Object] The raw response from the API call
+        # @return [Object] The processed response object
+        # @raise [Durable::Llm::APIError] If the response indicates an API error
+        # @raise [NotImplementedError] Subclasses must implement this method
         def handle_response(response)
           raise NotImplementedError, 'Subclasses must implement handle_response'
         end
@@ -63,3 +143,5 @@ module Durable
     end
   end
 end
+
+# Copyright (c) 2025 Durable Programming, LLC. All rights reserved.

data/lib/durable/llm/providers/cohere.rb

@@ -1,11 +1,20 @@
+# frozen_string_literal: true
+
+# This file implements the Cohere provider for accessing Cohere's language models through their API.
+
 require 'faraday'
 require 'json'
 require 'durable/llm/errors'
 require 'durable/llm/providers/base'
+require 'event_stream_parser'
 
 module Durable
   module Llm
     module Providers
+      # Cohere provider for accessing Cohere's language models
+      #
+      # This class provides completion, embedding, and streaming capabilities
+      # for Cohere's API, including proper error handling and response normalization.
       class Cohere < Durable::Llm::Providers::Base
         BASE_URL = 'https://api.cohere.ai/v2'
 
@@ -16,7 +25,7 @@ module Durable
         attr_accessor :api_key
 
         def initialize(api_key: nil)
-          @api_key = api_key || default_api_key
+          super(api_key: api_key)
           @conn = Faraday.new(url: BASE_URL) do |faraday|
             faraday.request :json
             faraday.response :json
@@ -34,10 +43,37 @@ module Durable
           handle_response(response)
         end
 
+        def stream(options)
+          options[:stream] = true
+
+          response = @conn.post('chat') do |req|
+            req.headers['Authorization'] = "Bearer #{@api_key}"
+            req.headers['Accept'] = 'text/event-stream'
+            req.body = options
+
+            user_proc = proc do |chunk, _size, _total|
+              yield CohereStreamResponse.new(chunk)
+            end
+
+            req.options.on_data = to_json_stream(user_proc: user_proc)
+          end
+
+          handle_response(response)
+        end
+
+        def embedding(model:, input:, **options)
+          response = @conn.post('embed') do |req|
+            req.headers['Authorization'] = "Bearer #{@api_key}"
+            req.headers['Content-Type'] = 'application/json'
+            req.body = { model: model, texts: Array(input), input_type: 'search_document', **options }
+          end
+
+          handle_response(response, CohereEmbeddingResponse)
+        end
+
         def models
-          response = @conn.get('models') do |req|
+          response = @conn.get('../v1/models') do |req|
             req.headers['Authorization'] = "Bearer #{@api_key}"
-            req.headers['OpenAI-Organization'] = @organization if @organization
           end
 
           data = handle_response(response).raw_response
@@ -45,28 +81,69 @@ module Durable
         end
 
         def self.stream?
-          false
+          true
         end
 
         private
 
-        def handle_response(response)
+        # CODE-FROM: ruby-openai @ https://github.com/alexrudall/ruby-openai/blob/main/lib/openai/http.rb
+        # MIT License: https://github.com/alexrudall/ruby-openai/blob/main/LICENSE.md
+        # Given a proc, returns an outer proc that can be used to iterate over a JSON stream of chunks.
+        # For each chunk, the inner user_proc is called giving it the JSON object. The JSON object could
+        # be a data object or an error object as described in the Cohere API documentation.
+        #
+        # @param user_proc [Proc] The inner proc to call for each JSON object in the chunk.
+        # @return [Proc] An outer proc that iterates over a raw stream, converting it to JSON.
+        def to_json_stream(user_proc:)
+          parser = EventStreamParser::Parser.new
+
+          proc do |chunk, _bytes, env|
+            if env && env.status != 200
+              raise_error = Faraday::Response::RaiseError.new
+              raise_error.on_complete(env.merge(body: try_parse_json(chunk)))
+            end
+
+            parser.feed(chunk) do |_type, data|
+              user_proc.call(JSON.parse(data)) unless data == '[DONE]'
+            end
+          end
+        end
+
+        def try_parse_json(maybe_json)
+          JSON.parse(maybe_json)
+        rescue JSON::ParserError
+          maybe_json
+        end
+
+        # END-CODE-FROM
+
+        def handle_response(response, response_class = CohereResponse)
           case response.status
           when 200..299
-            CohereResponse.new(response.body)
+            response_class.new(response.body)
           when 401
-            raise Durable::Llm::AuthenticationError, response.body['message']
+            raise Durable::Llm::AuthenticationError, parse_error_message(response)
           when 429
-            raise Durable::Llm::RateLimitError, response.body['message']
+            raise Durable::Llm::RateLimitError, parse_error_message(response)
           when 400..499
-            raise Durable::Llm::InvalidRequestError, response.body['message']
+            raise Durable::Llm::InvalidRequestError, parse_error_message(response)
           when 500..599
-            raise Durable::Llm::ServerError, response.body['message']
+            raise Durable::Llm::ServerError, parse_error_message(response)
           else
             raise Durable::Llm::APIError, "Unexpected response code: #{response.status}"
           end
         end
 
+        def parse_error_message(response)
+          body = begin
+            JSON.parse(response.body)
+          rescue StandardError
+            nil
+          end
+          message = body&.dig('message') || response.body
+          "#{response.status} Error: #{message}"
+        end
+
         class CohereResponse
           attr_reader :raw_response
 
@@ -75,7 +152,7 @@ module Durable
           end
 
           def choices
-            [@raw_response.dig('message', 'content')].flatten.map { |generation| CohereChoice.new(generation) }
+            @raw_response.dig('message', 'content')&.map { |generation| CohereChoice.new(generation) } || []
           end
 
           def to_s
@@ -94,7 +171,57 @@ module Durable
             @text
           end
         end
+
+        class CohereEmbeddingResponse
+          attr_reader :embedding
+
+          def initialize(data)
+            @embedding = data.dig('embeddings', 'float', 0)
+          end
+
+          def to_a
+            @embedding
+          end
+        end
+
+        class CohereStreamResponse
+          attr_reader :choices
+
+          def initialize(parsed)
+            @choices = [CohereStreamChoice.new(parsed['delta'])]
+          end
+
+          def to_s
+            @choices.map(&:to_s).join(' ')
+          end
+        end
+
+        class CohereStreamChoice
+          attr_reader :delta
+
+          def initialize(delta)
+            @delta = CohereStreamDelta.new(delta)
+          end
+
+          def to_s
+            @delta.to_s
+          end
+        end
+
+        class CohereStreamDelta
+          attr_reader :text
+
+          def initialize(delta)
+            @text = delta['text']
+          end
+
+          def to_s
+            @text || ''
+          end
+        end
       end
     end
   end
 end
+
+# Copyright (c) 2025 Durable Programming, LLC. All rights reserved.