llm.rb 4.1.0 → 4.2.0

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

@@ -10,10 +10,10 @@ class LLM::Anthropic
   #   require "llm"
   #
   #   llm = LLM.anthropic(key: ENV["KEY"])
-  #   bot = LLM::Bot.new(llm)
+  #   ses = LLM::Session.new(llm)
   #   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" }
+  #   ses.talk ["Tell me about this PDF", file]
+  #   ses.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class Files
     ##
     # Returns a new Files object
@@ -38,8 +38,9 @@ class LLM::Anthropic
     def all(**params)
       query = URI.encode_www_form(params)
       req = Net::HTTP::Get.new("/v1/files?#{query}", headers)
-      res = execute(request: req)
-      ResponseAdapter.adapt(res, type: :enumerable)
+      res, span = execute(request: req, operation: "request")
+      res = ResponseAdapter.adapt(res, type: :enumerable)
+      finish_trace(operation: "request", res:, span:)
     end
 
     ##
@@ -57,8 +58,9 @@ class LLM::Anthropic
       req = Net::HTTP::Post.new("/v1/files", headers)
       req["content-type"] = multi.content_type
       set_body_stream(req, multi.body)
-      res = execute(request: req)
-      ResponseAdapter.adapt(res, type: :file)
+      res, span = execute(request: req, operation: "request")
+      res = ResponseAdapter.adapt(res, type: :file)
+      finish_trace(operation: "request", res:, span:)
     end
 
     ##
@@ -76,8 +78,9 @@ class LLM::Anthropic
       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)
-      ResponseAdapter.adapt(res, type: :file)
+      res, span = execute(request: req, operation: "request")
+      res = ResponseAdapter.adapt(res, type: :file)
+      finish_trace(operation: "request", res:, span:)
     end
 
     ##
@@ -95,8 +98,9 @@ class LLM::Anthropic
       query = URI.encode_www_form(params)
       file_id = file.respond_to?(:id) ? file.id : file
       req = Net::HTTP::Get.new("/v1/files/#{file_id}?#{query}", headers)
-      res = execute(request: req)
-      ResponseAdapter.adapt(res, type: :file)
+      res, span = execute(request: req, operation: "request")
+      res = ResponseAdapter.adapt(res, type: :file)
+      finish_trace(operation: "request", res:, span:)
     end
     alias_method :retrieve_metadata, :get_metadata
 
@@ -113,8 +117,9 @@ class LLM::Anthropic
     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::Response.new(res)
+      res, span = execute(request: req, operation: "request")
+      res = LLM::Response.new(res)
+      finish_trace(operation: "request", res:, span:)
     end
 
     ##
@@ -137,8 +142,9 @@ class LLM::Anthropic
       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.new(res).tap { _1.define_singleton_method(:file) { io } }
+      res, span = execute(request: req, operation: "request") { |res| res.read_body { |chunk| io << chunk } }
+      res = LLM::Response.new(res).tap { _1.define_singleton_method(:file) { io } }
+      finish_trace(operation: "request", res:, span:)
     end
 
     private
@@ -147,7 +153,7 @@ class LLM::Anthropic
       @provider.instance_variable_get(:@key)
     end
 
-    [:headers, :execute, :set_body_stream].each do |m|
+    [:headers, :execute, :set_body_stream, :finish_trace].each do |m|
       define_method(m) { |*args, **kwargs, &b| @provider.send(m, *args, **kwargs, &b) }
     end
   end

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

@@ -40,13 +40,14 @@ class LLM::Anthropic
     def all(**params)
       query = URI.encode_www_form(params)
       req = Net::HTTP::Get.new("/v1/models?#{query}", headers)
-      res = execute(request: req)
-      ResponseAdapter.adapt(res, type: :enumerable)
+      res, span = execute(request: req, operation: "request")
+      res = ResponseAdapter.adapt(res, type: :enumerable)
+      finish_trace(operation: "request", res:, span:)
     end
 
     private
 
-    [:headers, :execute].each do |m|
+    [:headers, :execute, :finish_trace].each do |m|
       define_method(m) { |*args, **kwargs, &b| @provider.send(m, *args, **kwargs, &b) }
     end
   end

data/lib/llm/providers/anthropic.rb

@@ -10,9 +10,9 @@ module LLM
   #   require "llm"
   #
   #   llm = LLM.anthropic(key: ENV["KEY"])
-  #   bot = LLM::Bot.new(llm)
-  #   bot.chat ["Tell me about this photo", File.open("/images/dog.jpg", "rb")]
-  #   bot.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
+  #   ses = LLM::Session.new(llm)
+  #   ses.talk ["Tell me about this photo", ses.local_file("/images/photo.png")]
+  #   ses.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class Anthropic < Provider
     require_relative "anthropic/error_handler"
     require_relative "anthropic/request_adapter"
@@ -43,9 +43,10 @@ module LLM
     def complete(prompt, params = {})
       params, stream, tools, role = normalize_complete_params(params)
       req = build_complete_request(prompt, params, role)
-      res = execute(request: req, stream: stream)
-      ResponseAdapter.adapt(res, type: :completion)
+      res, span = execute(request: req, stream: stream, operation: "chat", model: params[:model])
+      res = ResponseAdapter.adapt(res, type: :completion)
         .extend(Module.new { define_method(:__tools__) { tools } })
+      finish_trace(operation: "chat", model: params[:model], res:, span:)
     end
 
     ##

data/lib/llm/providers/deepseek.rb

@@ -14,9 +14,9 @@ module LLM
   #   require "llm"
   #
   #   llm = LLM.deepseek(key: ENV["KEY"])
-  #   bot = LLM::Bot.new(llm)
-  #   bot.chat ["Tell me about this photo", File.open("/images/cat.jpg", "rb")]
-  #   bot.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
+  #   ses = LLM::Session.new(llm)
+  #   ses.talk ["Tell me about this photo", ses.local_file("/images/photo.png")]
+  #   ses.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class DeepSeek < OpenAI
     require_relative "deepseek/request_adapter"
     include DeepSeek::RequestAdapter

data/lib/llm/providers/gemini/error_handler.rb

@@ -10,10 +10,21 @@ class LLM::Gemini
     attr_reader :res
 
     ##
+    # @return [Object, nil]
+    #  The span
+    attr_reader :span
+
+    ##
+    # @param [LLM::Tracer] tracer
+    #  The tracer
+    # @param [Object, nil] span
+    #  The span
     # @param [Net::HTTPResponse] res
     #  The response from the server
     # @return [LLM::Gemini::ErrorHandler]
-    def initialize(res)
+    def initialize(tracer, span, res)
+      @tracer = tracer
+      @span = span
       @res = res
     end
 
@@ -21,27 +32,38 @@ class LLM::Gemini
     # @raise [LLM::Error]
     #  Raises a subclass of {LLM::Error LLM::Error}
     def raise_error!
+      ex = error
+      @tracer.on_request_error(ex:, span:)
+    ensure
+      raise(ex)
+    end
+
+    private
+
+    ##
+    # @return [LLM::Object]
+    def body
+      @body ||= LLM.json.load(res.body)
+    end
+
+    ##
+    # @return [LLM::Error]
+    def error
       case res
       when Net::HTTPServerError
-        raise LLM::ServerError.new { _1.response = res }, "Server error"
+        LLM::ServerError.new("Server error").tap { _1.response = res }
       when Net::HTTPBadRequest
         reason = body.dig("error", "details", 0, "reason")
         if reason == "API_KEY_INVALID"
-          raise LLM::UnauthorizedError.new { _1.response = res }, "Authentication error"
+          LLM::UnauthorizedError.new("Authentication error").tap { _1.response = res }
         else
-          raise LLM::Error.new { _1.response = res }, "Unexpected response"
+          LLM::Error.new("Unexpected response").tap { _1.response = res }
         end
       when Net::HTTPTooManyRequests
-        raise LLM::RateLimitError.new { _1.response = res }, "Too many requests"
+        LLM::RateLimitError.new("Too many requests").tap { _1.response = res }
       else
-        raise LLM::Error.new { _1.response = res }, "Unexpected response"
+        LLM::Error.new("Unexpected response").tap { _1.response = res }
       end
     end
-
-    private
-
-    def body
-      @body ||= LLM.json.load(res.body)
-    end
   end
 end

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

@@ -18,10 +18,10 @@ class LLM::Gemini
   #   require "llm"
   #
   #   llm = LLM.gemini(key: ENV["KEY"])
-  #   bot = LLM::Bot.new(llm)
+  #   ses = LLM::Session.new(llm)
   #   file = llm.files.create(file: "/audio/haiku.mp3")
-  #   bot.chat ["Tell me about this file", file]
-  #   bot.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
+  #   ses.talk ["Tell me about this file", file]
+  #   ses.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class Files
     ##
     # Returns a new Files object
@@ -46,8 +46,9 @@ class LLM::Gemini
     def all(**params)
       query = URI.encode_www_form(params.merge!(key: key))
       req = Net::HTTP::Get.new("/v1beta/files?#{query}", headers)
-      res = execute(request: req)
-      ResponseAdapter.adapt(res, type: :files)
+      res, span = execute(request: req, operation: "request")
+      res = ResponseAdapter.adapt(res, type: :files)
+      finish_trace(operation: "request", res:, span:)
     end
 
     ##
@@ -68,8 +69,9 @@ class LLM::Gemini
       req["X-Goog-Upload-Command"] = "upload, finalize"
       file.with_io do |io|
         set_body_stream(req, io)
-        res = execute(request: req)
-        ResponseAdapter.adapt(res, type: :file)
+        res, span = execute(request: req, operation: "request")
+        res = ResponseAdapter.adapt(res, type: :file)
+        finish_trace(operation: "request", res:, span:)
       end
     end
 
@@ -88,8 +90,9 @@ class LLM::Gemini
       file_id = file.respond_to?(:name) ? file.name : file.to_s
       query = URI.encode_www_form(params.merge!(key: key))
       req = Net::HTTP::Get.new("/v1beta/#{file_id}?#{query}", headers)
-      res = execute(request: req)
-      ResponseAdapter.adapt(res, type: :file)
+      res, span = execute(request: req, operation: "request")
+      res = ResponseAdapter.adapt(res, type: :file)
+      finish_trace(operation: "request", res:, span:)
     end
 
     ##
@@ -106,8 +109,9 @@ class LLM::Gemini
       file_id = file.respond_to?(:name) ? file.name : file.to_s
       query = URI.encode_www_form(params.merge!(key: key))
       req = Net::HTTP::Delete.new("/v1beta/#{file_id}?#{query}", headers)
-      res = execute(request: req)
-      LLM::Response.new(res)
+      res, span = execute(request: req, operation: "request")
+      res = LLM::Response.new(res)
+      finish_trace(operation: "request", res:, span:)
     end
 
     ##
@@ -128,7 +132,8 @@ class LLM::Gemini
       req["X-Goog-Upload-Header-Content-Length"] = file.bytesize
       req["X-Goog-Upload-Header-Content-Type"] = file.mime_type
       req.body = LLM.json.dump(file: {display_name: File.basename(file.path)})
-      res = execute(request: req)
+      res, span = execute(request: req, operation: "request")
+      finish_trace(operation: "request", res: LLM::Response.new(res), span:)
       res["x-goog-upload-url"]
     end
 
@@ -136,7 +141,7 @@ class LLM::Gemini
       @provider.instance_variable_get(:@key)
     end
 
-    [:headers, :execute, :set_body_stream].each do |m|
+    [:headers, :execute, :set_body_stream, :finish_trace].each do |m|
       define_method(m) { |*args, **kwargs, &b| @provider.send(m, *args, **kwargs, &b) }
     end
   end

data/lib/llm/providers/gemini/images.rb

@@ -51,8 +51,9 @@ class LLM::Gemini
         instances: [{prompt:}]
       })
       req.body = body
-      res = execute(request: req)
-      ResponseAdapter.adapt(res, type: :image)
+      res, span = execute(request: req, operation: "request")
+      res = ResponseAdapter.adapt(res, type: :image)
+      finish_trace(operation: "request", model:, res:, span:)
     end
 
     ##
@@ -89,7 +90,7 @@ class LLM::Gemini
       @provider.instance_variable_get(:@key)
     end
 
-    [:headers, :execute, :set_body_stream].each do |m|
+    [:headers, :execute, :set_body_stream, :finish_trace].each do |m|
       define_method(m) { |*args, **kwargs, &b| @provider.send(m, *args, **kwargs, &b) }
     end
   end

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

@@ -42,8 +42,9 @@ class LLM::Gemini
     def all(**params)
       query = URI.encode_www_form(params.merge!(key: key))
       req = Net::HTTP::Get.new("/v1beta/models?#{query}", headers)
-      res = execute(request: req)
-      ResponseAdapter.adapt(res, type: :models)
+      res, span = execute(request: req, operation: "request")
+      res = ResponseAdapter.adapt(res, type: :models)
+      finish_trace(operation: "request", res:, span:)
     end
 
     private
@@ -52,7 +53,7 @@ class LLM::Gemini
       @provider.instance_variable_get(:@key)
     end
 
-    [:headers, :execute].each do |m|
+    [:headers, :execute, :finish_trace].each do |m|
       define_method(m) { |*args, **kwargs, &b| @provider.send(m, *args, **kwargs, &b) }
     end
   end

data/lib/llm/providers/gemini.rb

@@ -14,9 +14,9 @@ module LLM
   #   require "llm"
   #
   #   llm = LLM.gemini(key: ENV["KEY"])
-  #   bot = LLM::Bot.new(llm)
-  #   bot.chat ["Tell me about this photo", File.open("/images/horse.jpg", "rb")]
-  #   bot.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
+  #   ses = LLM::Session.new(llm)
+  #   ses.talk ["Tell me about this photo", ses.local_file("/images/photo.png")]
+  #   ses.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class Gemini < Provider
     require_relative "gemini/error_handler"
     require_relative "gemini/request_adapter"
@@ -49,8 +49,9 @@ module LLM
       path = ["/v1beta/models/#{model}", "embedContent?key=#{@key}"].join(":")
       req = Net::HTTP::Post.new(path, headers)
       req.body = LLM.json.dump({content: {parts: [{text: input}]}})
-      res = execute(request: req)
-      ResponseAdapter.adapt(res, type: :embedding)
+      res, span = execute(request: req, operation: "embeddings", model:)
+      res = ResponseAdapter.adapt(res, type: :embedding)
+      finish_trace(operation: "embeddings", model:, res:, span:)
     end
 
     ##
@@ -66,9 +67,10 @@ module LLM
     def complete(prompt, params = {})
       params, stream, tools, role, model = normalize_complete_params(params)
       req = build_complete_request(prompt, params, role, model, stream)
-      res = execute(request: req, stream: stream)
-      ResponseAdapter.adapt(res, type: :completion)
+      res, span = execute(request: req, stream: stream, operation: "chat", model:)
+      res = ResponseAdapter.adapt(res, type: :completion)
         .extend(Module.new { define_method(:__tools__) { tools } })
+      finish_trace(operation: "chat", model:, res:, span:)
     end
 
     ##

data/lib/llm/providers/llamacpp.rb

@@ -16,9 +16,9 @@ module LLM
   #   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" }
+  #   ses = LLM::Session.new(llm)
+  #   ses.talk ["Tell me about this photo", ses.local_file("/images/photo.png")]
+  #   ses.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class LlamaCpp < OpenAI
     ##
     # @param (see LLM::Provider#initialize)

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

@@ -10,10 +10,21 @@ class LLM::Ollama
     attr_reader :res
 
     ##
+    # @return [Object, nil]
+    #  The span
+    attr_reader :span
+
+    ##
+    # @param [LLM::Tracer] tracer
+    #  The tracer
+    # @param [Object, nil] span
+    #  The span
     # @param [Net::HTTPResponse] res
     #  The response from the server
-    # @return [LLM::OpenAI::ErrorHandler]
-    def initialize(res)
+    # @return [LLM::Ollama::ErrorHandler]
+    def initialize(tracer, span, res)
+      @tracer = tracer
+      @span = span
       @res = res
     end
 
@@ -21,15 +32,26 @@ class LLM::Ollama
     # @raise [LLM::Error]
     #  Raises a subclass of {LLM::Error LLM::Error}
     def raise_error!
+      ex = error
+      @tracer.on_request_error(ex:, span:)
+    ensure
+      raise(ex)
+    end
+
+    private
+
+    ##
+    # @return [LLM::Error]
+    def error
       case res
       when Net::HTTPServerError
-        raise LLM::ServerError.new { _1.response = res }, "Server error"
+        LLM::ServerError.new("Server error").tap { _1.response = res }
       when Net::HTTPUnauthorized
-        raise LLM::UnauthorizedError.new { _1.response = res }, "Authentication error"
+        LLM::UnauthorizedError.new("Authentication error").tap { _1.response = res }
       when Net::HTTPTooManyRequests
-        raise LLM::RateLimitError.new { _1.response = res }, "Too many requests"
+        LLM::RateLimitError.new("Too many requests").tap { _1.response = res }
       else
-        raise LLM::Error.new { _1.response = res }, "Unexpected response"
+        LLM::Error.new("Unexpected response").tap { _1.response = res }
       end
     end
   end

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

@@ -43,13 +43,14 @@ class LLM::Ollama
     def all(**params)
       query = URI.encode_www_form(params)
       req = Net::HTTP::Get.new("/api/tags?#{query}", headers)
-      res = execute(request: req)
-      LLM::Response.new(res)
+      res, span = execute(request: req, operation: "request")
+      res = LLM::Response.new(res)
+      finish_trace(operation: "request", res:, span:)
     end
 
     private
 
-    [:headers, :execute].each do |m|
+    [:headers, :execute, :finish_trace].each do |m|
       define_method(m) { |*args, **kwargs, &b| @provider.send(m, *args, **kwargs, &b) }
     end
   end

data/lib/llm/providers/ollama.rb

@@ -12,9 +12,9 @@ module LLM
   #   require "llm"
   #
   #   llm = LLM.ollama(key: nil)
-  #   bot = LLM::Bot.new(llm, model: "llava")
-  #   bot.chat ["Tell me about this image", File.open("/images/parrot.png", "rb")]
-  #   bot.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
+  #   ses = LLM::Session.new(llm, model: "llava")
+  #   ses.talk ["Tell me about this image", ses.local_file("/images/photo.png")]
+  #   ses.messages.select(&:assistant?).each { print "[#{_1.role}]", _1.content, "\n" }
   class Ollama < Provider
     require_relative "ollama/error_handler"
     require_relative "ollama/request_adapter"
@@ -43,8 +43,9 @@ module LLM
       params   = {model:}.merge!(params)
       req      = Net::HTTP::Post.new("/v1/embeddings", headers)
       req.body = LLM.json.dump({input:}.merge!(params))
-      res      = execute(request: req)
-      ResponseAdapter.adapt(res, type: :embedding)
+      res, span = execute(request: req, operation: "embeddings", model:)
+      res = ResponseAdapter.adapt(res, type: :embedding)
+      finish_trace(operation: "embeddings", model:, res:, span:)
     end
 
     ##
@@ -60,9 +61,10 @@ module LLM
     def complete(prompt, params = {})
       params, stream, tools, role = normalize_complete_params(params)
       req = build_complete_request(prompt, params, role)
-      res = execute(request: req, stream: stream)
-      ResponseAdapter.adapt(res, type: :completion)
+      res, span = execute(request: req, stream: stream, operation: "chat", model: params[:model])
+      res = ResponseAdapter.adapt(res, type: :completion)
         .extend(Module.new { define_method(:__tools__) { tools } })
+      finish_trace(operation: "chat", model: params[:model], res:, span:)
     end
 
     ##

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

@@ -35,8 +35,9 @@ class LLM::OpenAI
       req = Net::HTTP::Post.new("/v1/audio/speech", headers)
       req.body = LLM.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.new(res).tap { _1.define_singleton_method(:audio) { io } }
+      res, span = execute(request: req, operation: "request") { _1.read_body { |chunk| io << chunk } }
+      res = LLM::Response.new(res).tap { _1.define_singleton_method(:audio) { io } }
+      finish_trace(operation: "request", model:, res:, span:)
     end
 
     ##
@@ -56,8 +57,9 @@ class LLM::OpenAI
       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.new(res)
+      res, span = execute(request: req, operation: "request")
+      res = LLM::Response.new(res)
+      finish_trace(operation: "request", model:, res:, span:)
     end
 
     ##
@@ -78,13 +80,14 @@ class LLM::OpenAI
       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.new(res)
+      res, span = execute(request: req, operation: "request")
+      res = LLM::Response.new(res)
+      finish_trace(operation: "request", model:, res:, span:)
     end
 
     private
 
-    [:headers, :execute, :set_body_stream].each do |m|
+    [:headers, :execute, :set_body_stream, :finish_trace].each do |m|
       define_method(m) { |*args, **kwargs, &b| @provider.send(m, *args, **kwargs, &b) }
     end
   end

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

@@ -10,10 +10,21 @@ class LLM::OpenAI
     attr_reader :res
 
     ##
+    # @return [Object, nil]
+    #  The span
+    attr_reader :span
+
+    ##
+    # @param [LLM::Tracer] tracer
+    #  The tracer
+    # @param [Object, nil] span
+    #  The span
     # @param [Net::HTTPResponse] res
     #  The response from the server
     # @return [LLM::OpenAI::ErrorHandler]
-    def initialize(res)
+    def initialize(tracer, span, res)
+      @tracer = tracer
+      @span = span
       @res = res
     end
 
@@ -21,36 +32,52 @@ class LLM::OpenAI
     # @raise [LLM::Error]
     #  Raises a subclass of {LLM::Error LLM::Error}
     def raise_error!
+      ex = error
+      @tracer.on_request_error(ex:, span:)
+    ensure
+      raise(ex)
+    end
+
+    private
+
+    ##
+    # @return [LLM::Object]
+    def body
+      @body ||= LLM.json.load(res.body)
+    end
+
+    ##
+    # @return [LLM::Error]
+    def error
       case res
       when Net::HTTPServerError
-        raise LLM::ServerError.new { _1.response = res }, "Server error"
+        LLM::ServerError.new("Server error").tap { _1.response = res }
       when Net::HTTPUnauthorized
-        raise LLM::UnauthorizedError.new { _1.response = res }, "Authentication error"
+        LLM::UnauthorizedError.new("Authentication error").tap { _1.response = res }
       when Net::HTTPTooManyRequests
-        raise LLM::RateLimitError.new { _1.response = res }, "Too many requests"
+        LLM::RateLimitError.new("Too many requests").tap { _1.response = res }
       else
         error = body["error"] || {}
         case error["type"]
         when "invalid_request_error" then handle_invalid_request(error)
-        when "server_error" then raise LLM::ServerError.new { _1.response = res }, error["message"]
-        else raise LLM::Error.new { _1.response = res }, error["message"] || "Unexpected response"
+        when "server_error"
+          LLM::ServerError.new(error["message"]).tap { _1.response = res }
+        else
+          LLM::Error.new(error["message"] || "Unexpected response").tap { _1.response = res }
         end
       end
     end
 
-    private
-
+    ##
+    # @param [Exception] error
+    # @return [LLM::Error]
     def handle_invalid_request(error)
       case error["code"]
       when "context_length_exceeded"
-        raise LLM::ContextWindowError.new { _1.response = res }, error["message"]
+        LLM::ContextWindowError.new(error["message"]).tap { _1.response = res }
       else
-        raise LLM::InvalidRequestError.new { _1.response = res }, error["message"]
+        LLM::InvalidRequestError.new(error["message"]).tap { _1.response = res }
       end
     end
-
-    def body
-      @body ||= LLM.json.load(res.body)
-    end
   end
 end