llm.rb 0.10.1 → 0.12.0

data/lib/llm/providers/gemini/{response_parser/completion_parser.rb → response/completion.rb}

@@ -1,30 +1,23 @@
 # frozen_string_literal: true
 
-module LLM::Gemini::ResponseParser
-  class CompletionParser
-    def initialize(body)
-      @body = LLM::Object.from_hash(body)
-    end
-
-    def format(response)
-      {
-        model:,
-        prompt_tokens:,
-        completion_tokens:,
-        total_tokens:,
-        choices: format_choices(response)
-      }
-    end
+module LLM::Gemini::Response
+  module Completion
+    def model = body.modelVersion
+    def prompt_tokens = body.usageMetadata.promptTokenCount
+    def completion_tokens = body.usageMetadata.candidatesTokenCount
+    def total_tokens = body.usageMetadata.totalTokenCount
+    def choices = format_choices
 
     private
 
-    def format_choices(response)
+    def format_choices
       candidates.map.with_index do |choice, index|
+        choice = LLM::Object.from_hash(choice)
         content = choice.content
         role, parts = content.role, content.parts
         text  = parts.filter_map { _1["text"] }.join
         tools = parts.filter_map { _1["functionCall"] }
-        extra = {index:, response:, tool_calls: format_tool_calls(tools), original_tool_calls: tools}
+        extra = {index:, response: self, tool_calls: format_tool_calls(tools), original_tool_calls: tools}
         LLM::Message.new(role, text, extra)
       end
     end
@@ -35,12 +28,5 @@ module LLM::Gemini::ResponseParser
         LLM::Object.new(function)
       end
     end
-
-    def body = @body
-    def model = body.modelVersion
-    def prompt_tokens = body.usageMetadata.promptTokenCount
-    def completion_tokens = body.usageMetadata.candidatesTokenCount
-    def total_tokens = body.usageMetadata.totalTokenCount
-    def candidates = body.candidates
   end
 end

data/lib/llm/providers/gemini/response/embedding.rb

@@ -0,0 +1,8 @@
+# frozen_string_literal: true
+
+module LLM::Gemini::Response
+  module Embedding
+    def model = "text-embedding-004"
+    def embeddings = body.dig("embedding", "values")
+  end
+end

data/lib/llm/providers/gemini/response/file.rb

@@ -0,0 +1,11 @@
+# frozen_string_literal: true
+
+module LLM::Gemini::Response
+  module File
+    def name = respond_to?(:file) ? file.name : body.name
+    def display_name = respond_to?(:file) ? file.displayName : body.displayName
+    def mime_type = respond_to?(:file) ? file.mimeType : body.mimeType
+    def uri = respond_to?(:file) ? file.uri : body.uri
+    def file? = true
+  end
+end

data/lib/llm/providers/gemini/response/image.rb

@@ -0,0 +1,26 @@
+# frozen_string_literal: true
+
+module LLM::Gemini::Response
+  module Image
+    ##
+    # @return [Array<StringIO>]
+    def images
+      candidates.flat_map do |candidate|
+        parts = candidate["content"]["parts"]
+        parts.filter_map do
+          data = _1.dig(:inlineData, :data)
+          next unless data
+          StringIO.new(data.unpack1("m0"))
+        end
+      end
+    end
+
+    ##
+    # Returns one or more image URLs, or an empty array
+    # @note
+    #  Gemini's image generation API does not return URLs, so this method
+    #  will always return an empty array.
+    # @return [Array<String>]
+    def urls = []
+  end
+end

data/lib/llm/providers/gemini.rb

@@ -3,36 +3,26 @@
 module LLM
   ##
   # The Gemini class implements a provider for
-  # [Gemini](https://ai.google.dev/).
+  # [Gemini](https://ai.google.dev/). The Gemini provider
+  # can accept multiple inputs (text, images, audio, and video).
+  # The inputs can be provided inline via the prompt for files
+  # under 20MB or via the Gemini Files API for files
+  # that are over 20MB.
   #
-  # The Gemini provider can accept multiple inputs (text, images,
-  # audio, and video). The inputs can be provided inline via the
-  # prompt for files under 20MB or via the Gemini Files API for
-  # files that are over 20MB
-  #
-  # @example example #1
+  # @example
   #   #!/usr/bin/env ruby
   #   require "llm"
   #
-  #   llm = LLM.gemini(ENV["KEY"])
+  #   llm = LLM.gemini(key: ENV["KEY"])
   #   bot = LLM::Bot.new(llm)
-  #   bot.chat LLM.File("/images/capybara.png")
-  #   bot.chat "Describe the image"
-  #   bot.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
-  #
-  # @example example #2
-  #   #!/usr/bin/env ruby
-  #   require "llm"
-  #
-  #   llm = LLM.gemini(ENV["KEY"])
-  #   bot = LLM::Bot.new(llm)
-  #   bot.chat ["Describe the image", LLM::File("/images/capybara.png")]
+  #   bot.chat ["Tell me about this photo", File.open("/images/horse.jpg", "rb")]
   #   bot.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class Gemini < Provider
+    require_relative "gemini/response/embedding"
+    require_relative "gemini/response/completion"
     require_relative "gemini/error_handler"
     require_relative "gemini/format"
     require_relative "gemini/stream_parser"
-    require_relative "gemini/response_parser"
     require_relative "gemini/models"
     require_relative "gemini/images"
     require_relative "gemini/files"
@@ -54,14 +44,14 @@ module LLM
     # @param model (see LLM::Provider#embed)
     # @param params (see LLM::Provider#embed)
     # @raise (see LLM::Provider#request)
-    # @return (see LLM::Provider#embed)
+    # @return [LLM::Response]
     def embed(input, model: "text-embedding-004", **params)
       model = model.respond_to?(:id) ? model.id : model
       path = ["/v1beta/models/#{model}", "embedContent?key=#{@key}"].join(":")
       req = Net::HTTP::Post.new(path, headers)
       req.body = JSON.dump({content: {parts: [{text: input}]}})
       res = execute(request: req)
-      Response::Embedding.new(res).extend(response_parser)
+      LLM::Response.new(res).extend(LLM::Gemini::Response::Embedding)
     end
 
     ##
@@ -71,9 +61,9 @@ module LLM
     # @param params (see LLM::Provider#complete)
     # @example (see LLM::Provider#complete)
     # @raise (see LLM::Provider#request)
-    # @raise [LLM::Error::PromptError]
+    # @raise [LLM::PromptError]
     #  When given an object a provider does not understand
-    # @return (see LLM::Provider#complete)
+    # @return [LLM::Response]
     def complete(prompt, params = {})
       params = {role: :user, model: default_model}.merge!(params)
       params = [params, format_schema(params), format_tools(params)].inject({}, &:merge!).compact
@@ -86,12 +76,13 @@ module LLM
       body = JSON.dump({contents: format(messages)}.merge!(params))
       set_body_stream(req, StringIO.new(body))
       res = execute(request: req, stream:)
-      Response::Completion.new(res).extend(response_parser)
+      LLM::Response.new(res).extend(LLM::Gemini::Response::Completion)
     end
 
     ##
     # Provides an interface to Gemini's audio API
     # @see https://ai.google.dev/gemini-api/docs/audio Gemini docs
+    # @return [LLM::Gemini::Audio]
     def audio
       LLM::Gemini::Audio.new(self)
     end
@@ -107,6 +98,7 @@ module LLM
     ##
     # Provides an interface to Gemini's file management API
     # @see https://ai.google.dev/gemini-api/docs/files Gemini docs
+    # @return [LLM::Gemini::Files]
     def files
       LLM::Gemini::Files.new(self)
     end
@@ -114,6 +106,7 @@ module LLM
     ##
     # Provides an interface to Gemini's models API
     # @see https://ai.google.dev/gemini-api/docs/models Gemini docs
+    # @return [LLM::Gemini::Models]
     def models
       LLM::Gemini::Models.new(self)
     end
@@ -140,10 +133,6 @@ module LLM
       )
     end
 
-    def response_parser
-      LLM::Gemini::ResponseParser
-    end
-
     def stream_parser
       LLM::Gemini::StreamParser
     end

data/lib/llm/providers/llamacpp.rb

@@ -7,7 +7,18 @@ module LLM
   # The LlamaCpp class implements a provider for
   # [llama.cpp](https://github.com/ggml-org/llama.cpp)
   # through the OpenAI-compatible API provided by the
-  # llama-server binary.
+  # llama-server binary. Similar to the ollama provider,
+  # this provider supports a wide range of models and
+  # is straightforward to run on your own hardware.
+  #
+  # @example
+  #   #!/usr/bin/env ruby
+  #   require "llm"
+  #
+  #   llm = LLM.llamacpp(key: nil)
+  #   bot = LLM::Bot.new(llm)
+  #   bot.chat ["Tell me about this photo", File.open("/images/frog.jpg", "rb")]
+  #   bot.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class LlamaCpp < OpenAI
     ##
     # @param (see LLM::Provider#initialize)
@@ -46,6 +57,12 @@ module LLM
       raise NotImplementedError
     end
 
+    ##
+    # @raise [NotImplementedError]
+    def vector_stores
+      raise NotImplementedError
+    end
+
     ##
     # Returns the default model for chat completions
     # @see https://ollama.com/library/qwen3 qwen3

data/lib/llm/providers/ollama/format/completion_format.rb

@@ -28,13 +28,16 @@ module LLM::Ollama::Format
 
     def format_content(content)
       case content
+      when File
+        content.close unless content.closed?
+        format_content(LLM.File(content.path))
       when LLM::File
         if content.image?
           {content: "This message has an image associated with it", images: [content.to_b64]}
         else
-          raise LLM::Error::PromptError, "The given object (an instance of #{content.class}) " \
-                                         "is not an image, and therefore not supported by the " \
-                                         "Ollama API"
+          raise LLM::PromptError, "The given object (an instance of #{content.class}) " \
+                                  "is not an image, and therefore not supported by the " \
+                                  "Ollama API"
         end
       when String
         {content:}
@@ -43,8 +46,8 @@ module LLM::Ollama::Format
       when LLM::Function::Return
         throw(:abort, {role: "tool", tool_call_id: content.id, content: JSON.dump(content.value)})
       else
-        raise LLM::Error::PromptError, "The given object (an instance of #{content.class}) " \
-                                       "is not supported by the Ollama API"
+        raise LLM::PromptError, "The given object (an instance of #{content.class}) " \
+                                "is not supported by the Ollama API"
       end
     end
 

data/lib/llm/providers/ollama/models.rb

@@ -39,18 +39,12 @@ class LLM::Ollama
     # @see https://ollama.com/library Ollama library
     # @param [Hash] params Other parameters (see Ollama docs)
     # @raise (see LLM::Provider#request)
-    # @return [LLM::Response::ModelList]
+    # @return [LLM::Response]
     def all(**params)
       query = URI.encode_www_form(params)
       req = Net::HTTP::Get.new("/api/tags?#{query}", headers)
       res = execute(request: req)
-      LLM::Response::ModelList.new(res).tap { |modellist|
-        models = modellist.body["models"].map do |model|
-          model = model.transform_keys { snakecase(_1) }
-          LLM::Model.from_hash(model).tap { _1.provider = @provider }
-        end
-        modellist.models = models
-      }
+      LLM::Response.new(res)
     end
 
     private

data/lib/llm/providers/ollama/response/completion.rb

@@ -0,0 +1,28 @@
+# frozen_string_literal: true
+
+module LLM::Ollama::Response
+  module Completion
+    def model = body.model
+    def prompt_tokens = body.prompt_eval_count || 0
+    def completion_tokens = body.eval_count || 0
+    def total_tokens = prompt_tokens + completion_tokens
+    def message = body.message
+    def choices = [format_choices]
+
+    private
+
+    def format_choices
+      role, content, calls = message.to_h.values_at("role", "content", "tool_calls")
+      extra = {response: self, tool_calls: format_tool_calls(calls)}
+      LLM::Message.new(role, content, extra)
+    end
+
+    def format_tool_calls(tools)
+      return [] unless tools
+      tools.filter_map do |tool|
+        next unless tool["function"]
+        LLM::Object.new(tool["function"])
+      end
+    end
+  end
+end

data/lib/llm/providers/ollama/response/embedding.rb

@@ -0,0 +1,9 @@
+# frozen_string_literal: true
+
+module LLM::Ollama::Response
+  module Embedding
+    def embeddings = data.map { _1["embedding"] }
+    def prompt_tokens = body.dig("usage", "prompt_tokens") || 0
+    def total_tokens = body.dig("usage", "total_tokens") || 0
+  end
+end

data/lib/llm/providers/ollama.rb

@@ -2,27 +2,25 @@
 
 module LLM
   ##
-  # The Ollama class implements a provider for [Ollama](https://ollama.ai/).
-  #
-  # This provider supports a wide range of models, it is relatively
-  # straight forward to run on your own hardware, and includes multi-modal
-  # models that can process images and text. See the example for a demonstration
-  # of a multi-modal model by the name `llava`
+  # The Ollama class implements a provider for [Ollama](https://ollama.ai/) –
+  # and the provider supports a wide range of models. It is straight forward
+  # to run on your own hardware, and there are a number of multi-modal models
+  # that can process both images and text.
   #
   # @example
   #   #!/usr/bin/env ruby
   #   require "llm"
   #
-  #   llm = LLM.ollama(nil)
+  #   llm = LLM.ollama(key: nil)
   #   bot = LLM::Bot.new(llm, model: "llava")
-  #   bot.chat LLM::File("/images/capybara.png")
-  #   bot.chat "Describe the image"
+  #   bot.chat ["Tell me about this image", File.open("/images/parrot.png", "rb")]
   #   bot.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class Ollama < Provider
+    require_relative "ollama/response/embedding"
+    require_relative "ollama/response/completion"
     require_relative "ollama/error_handler"
     require_relative "ollama/format"
     require_relative "ollama/stream_parser"
-    require_relative "ollama/response_parser"
     require_relative "ollama/models"
 
     include Format
@@ -41,13 +39,13 @@ module LLM
     # @param model (see LLM::Provider#embed)
     # @param params (see LLM::Provider#embed)
     # @raise (see LLM::Provider#request)
-    # @return (see LLM::Provider#embed)
+    # @return [LLM::Response]
     def embed(input, model: default_model, **params)
       params   = {model:}.merge!(params)
       req      = Net::HTTP::Post.new("/v1/embeddings", headers)
       req.body = JSON.dump({input:}.merge!(params))
       res      = execute(request: req)
-      Response::Embedding.new(res).extend(response_parser)
+      LLM::Response.new(res).extend(LLM::Ollama::Response::Embedding)
     end
 
     ##
@@ -57,9 +55,9 @@ module LLM
     # @param params (see LLM::Provider#complete)
     # @example (see LLM::Provider#complete)
     # @raise (see LLM::Provider#request)
-    # @raise [LLM::Error::PromptError]
+    # @raise [LLM::PromptError]
     #  When given an object a provider does not understand
-    # @return (see LLM::Provider#complete)
+    # @return [LLM::Response]
     def complete(prompt, params = {})
       params = {role: :user, model: default_model, stream: true}.merge!(params)
       params = [params, {format: params[:schema]}, format_tools(params)].inject({}, &:merge!).compact
@@ -70,7 +68,7 @@ module LLM
       body = JSON.dump({messages: [format(messages)].flatten}.merge!(params))
       set_body_stream(req, StringIO.new(body))
       res = execute(request: req, stream:)
-      Response::Completion.new(res).extend(response_parser)
+      LLM::Response.new(res).extend(LLM::Ollama::Response::Completion)
     end
 
     ##
@@ -104,10 +102,6 @@ module LLM
       )
     end
 
-    def response_parser
-      LLM::Ollama::ResponseParser
-    end
-
     def stream_parser
       LLM::Ollama::StreamParser
     end

data/lib/llm/providers/openai/audio.rb

@@ -5,7 +5,7 @@ class LLM::OpenAI
   # The {LLM::OpenAI::Audio LLM::OpenAI::Audio} class provides an audio
   # object for interacting with [OpenAI's audio API](https://platform.openai.com/docs/api-reference/audio/createSpeech).
   # @example
-  #   llm = LLM.openai(ENV["KEY"])
+  #   llm = LLM.openai(key: ENV["KEY"])
   #   res = llm.audio.create_speech(input: "A dog on a rocket to the moon")
   #   IO.copy_stream res.audio, "rocket.mp3"
   class Audio
@@ -20,7 +20,7 @@ class LLM::OpenAI
     ##
     # Create an audio track
     # @example
-    #   llm = LLM.openai(ENV["KEY"])
+    #   llm = LLM.openai(key: ENV["KEY"])
     #   res = llm.images.create_speech(input: "A dog on a rocket to the moon")
     #   File.binwrite("rocket.mp3", res.audio.string)
     # @see https://platform.openai.com/docs/api-reference/audio/createSpeech OpenAI docs
@@ -30,19 +30,19 @@ class LLM::OpenAI
     # @param [String] response_format The response format
     # @param [Hash] params Other parameters (see OpenAI docs)
     # @raise (see LLM::Provider#request)
-    # @return [LLM::Response::Audio]
+    # @return [LLM::Response]
     def create_speech(input:, voice: "alloy", model: "gpt-4o-mini-tts", response_format: "mp3", **params)
       req = Net::HTTP::Post.new("/v1/audio/speech", headers)
       req.body = JSON.dump({input:, voice:, model:, response_format:}.merge!(params))
       io = StringIO.new("".b)
       res = execute(request: req) { _1.read_body { |chunk| io << chunk } }
-      LLM::Response::Audio.new(res).tap { _1.audio = io }
+      LLM::Response.new(res).tap { _1.define_singleton_method(:audio) { io } }
     end
 
     ##
     # Create an audio transcription
     # @example
-    #   llm = LLM.openai(ENV["KEY"])
+    #   llm = LLM.openai(key: ENV["KEY"])
     #   res = llm.audio.create_transcription(file: "/audio/rocket.mp3")
     #   res.text # => "A dog on a rocket to the moon"
     # @see https://platform.openai.com/docs/api-reference/audio/createTranscription OpenAI docs
@@ -50,21 +50,21 @@ class LLM::OpenAI
     # @param [String] model The model to use
     # @param [Hash] params Other parameters (see OpenAI docs)
     # @raise (see LLM::Provider#request)
-    # @return [LLM::Response::AudioTranscription]
+    # @return [LLM::Response]
     def create_transcription(file:, model: "whisper-1", **params)
       multi = LLM::Multipart.new(params.merge!(file: LLM.File(file), model:))
       req = Net::HTTP::Post.new("/v1/audio/transcriptions", headers)
       req["content-type"] = multi.content_type
       set_body_stream(req, multi.body)
       res = execute(request: req)
-      LLM::Response::AudioTranscription.new(res).tap { _1.text = _1.body["text"] }
+      LLM::Response.new(res)
     end
 
     ##
     # Create an audio translation (in English)
     # @example
     #   # Arabic => English
-    #   llm = LLM.openai(ENV["KEY"])
+    #   llm = LLM.openai(key: ENV["KEY"])
     #   res = llm.audio.create_translation(file: "/audio/bismillah.mp3")
     #   res.text # => "In the name of Allah, the Beneficent, the Merciful."
     # @see https://platform.openai.com/docs/api-reference/audio/createTranslation OpenAI docs
@@ -72,14 +72,14 @@ class LLM::OpenAI
     # @param [String] model The model to use
     # @param [Hash] params Other parameters (see OpenAI docs)
     # @raise (see LLM::Provider#request)
-    # @return [LLM::Response::AudioTranslation]
+    # @return [LLM::Response]
     def create_translation(file:, model: "whisper-1", **params)
       multi = LLM::Multipart.new(params.merge!(file: LLM.File(file), model:))
       req = Net::HTTP::Post.new("/v1/audio/translations", headers)
       req["content-type"] = multi.content_type
       set_body_stream(req, multi.body)
       res = execute(request: req)
-      LLM::Response::AudioTranslation.new(res).tap { _1.text = _1.body["text"] }
+      LLM::Response.new(res)
     end
 
     private

data/lib/llm/providers/openai/files.rb

@@ -12,23 +12,14 @@ class LLM::OpenAI
   #   #!/usr/bin/env ruby
   #   require "llm"
   #
-  #   llm = LLM.openai(ENV["KEY"])
+  #   llm = LLM.openai(key: ENV["KEY"])
   #   bot = LLM::Bot.new(llm)
-  #   file = llm.files.create file: "/documents/freebsd.pdf"
-  #   bot.chat(file)
-  #   bot.chat("Describe the document")
-  #   bot.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
-  #
-  # @example example #2
-  #   #!/usr/bin/env ruby
-  #   require "llm"
-  #
-  #   llm = LLM.openai(ENV["KEY"])
-  #   bot = LLM::Bot.new(llm)
-  #   file = llm.files.create file: "/documents/openbsd.pdf"
-  #   bot.chat(["Describe the document I sent to you", file])
+  #   file = llm.files.create file: "/books/goodread.pdf"
+  #   bot.chat ["Tell me about this PDF", file]
   #   bot.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class Files
+    require_relative "response/file"
+
     ##
     # Returns a new Files object
     # @param provider [LLM::Provider]
@@ -40,7 +31,7 @@ class LLM::OpenAI
     ##
     # List all files
     # @example
-    #   llm = LLM.openai(ENV["KEY"])
+    #   llm = LLM.openai(key: ENV["KEY"])
     #   res = llm.files.all
     #   res.each do |file|
     #     print "id: ", file.id, "\n"
@@ -48,60 +39,57 @@ class LLM::OpenAI
     # @see https://platform.openai.com/docs/api-reference/files/list OpenAI docs
     # @param [Hash] params Other parameters (see OpenAI docs)
     # @raise (see LLM::Provider#request)
-    # @return [LLM::Response::FileList]
+    # @return [LLM::Response]
     def all(**params)
       query = URI.encode_www_form(params)
       req = Net::HTTP::Get.new("/v1/files?#{query}", headers)
       res = execute(request: req)
-      LLM::Response::FileList.new(res).tap { |filelist|
-        files = filelist.body["data"].map { LLM::Object.from_hash(_1) }
-        filelist.files = files
-      }
+      LLM::Response.new(res)
     end
 
     ##
     # Create a file
     # @example
-    #   llm = LLM.openai(ENV["KEY"])
+    #   llm = LLM.openai(key: ENV["KEY"])
     #   res = llm.files.create file: "/documents/haiku.txt"
     # @see https://platform.openai.com/docs/api-reference/files/create OpenAI docs
-    # @param [File] file The file
+    # @param [File, LLM::File, String] file The file
     # @param [String] purpose The purpose of the file (see OpenAI docs)
     # @param [Hash] params Other parameters (see OpenAI docs)
     # @raise (see LLM::Provider#request)
-    # @return [LLM::Response::File]
+    # @return [LLM::Response]
     def create(file:, purpose: "assistants", **params)
-      multi = LLM::Multipart.new(params.merge!(file:, purpose:))
+      multi = LLM::Multipart.new(params.merge!(file: LLM.File(file), purpose:))
       req = Net::HTTP::Post.new("/v1/files", headers)
       req["content-type"] = multi.content_type
       set_body_stream(req, multi.body)
       res = execute(request: req)
-      LLM::Response::File.new(res)
+      LLM::Response.new(res).extend(LLM::OpenAI::Response::File)
     end
 
     ##
     # Get a file
     # @example
-    #   llm = LLM.openai(ENV["KEY"])
+    #   llm = LLM.openai(key: ENV["KEY"])
     #   res = llm.files.get(file: "file-1234567890")
     #   print "id: ", res.id, "\n"
     # @see https://platform.openai.com/docs/api-reference/files/get OpenAI docs
     # @param [#id, #to_s] file The file ID
     # @param [Hash] params Other parameters (see OpenAI docs)
     # @raise (see LLM::Provider#request)
-    # @return [LLM::Response::File]
+    # @return [LLM::Response]
     def get(file:, **params)
       file_id = file.respond_to?(:id) ? file.id : file
       query = URI.encode_www_form(params)
       req = Net::HTTP::Get.new("/v1/files/#{file_id}?#{query}", headers)
       res = execute(request: req)
-      LLM::Response::File.new(res)
+      LLM::Response.new(res).extend(LLM::OpenAI::Response::File)
     end
 
     ##
     # Download the content of a file
     # @example
-    #   llm = LLM.openai(ENV["KEY"])
+    #   llm = LLM.openai(key: ENV["KEY"])
     #   res = llm.files.download(file: "file-1234567890")
     #   File.binwrite "haiku1.txt", res.file.read
     #   print res.file.read, "\n"
@@ -109,31 +97,31 @@ class LLM::OpenAI
     # @param [#id, #to_s] file The file ID
     # @param [Hash] params Other parameters (see OpenAI docs)
     # @raise (see LLM::Provider#request)
-    # @return [LLM::Response::DownloadFile]
+    # @return [LLM::Response]
     def download(file:, **params)
       query = URI.encode_www_form(params)
       file_id = file.respond_to?(:id) ? file.id : file
       req = Net::HTTP::Get.new("/v1/files/#{file_id}/content?#{query}", headers)
       io = StringIO.new("".b)
       res = execute(request: req) { |res| res.read_body { |chunk| io << chunk } }
-      LLM::Response::DownloadFile.new(res).tap { _1.file = io }
+      LLM::Response.new(res).tap { _1.define_singleton_method(:file) { io } }
     end
 
     ##
     # Delete a file
     # @example
-    #   llm = LLM.openai(ENV["KEY"])
+    #   llm = LLM.openai(key: ENV["KEY"])
     #   res = llm.files.delete(file: "file-1234567890")
     #   print res.deleted, "\n"
     # @see https://platform.openai.com/docs/api-reference/files/delete OpenAI docs
     # @param [#id, #to_s] file The file ID
     # @raise (see LLM::Provider#request)
-    # @return [LLM::Object] Response body
+    # @return [LLM::Response]
     def delete(file:)
       file_id = file.respond_to?(:id) ? file.id : file
       req = Net::HTTP::Delete.new("/v1/files/#{file_id}", headers)
       res = execute(request: req)
-      LLM::Object.from_hash JSON.parse(res.body)
+      LLM::Response.new(res)
     end
 
     private