llm.rb 8.0.0 → 9.0.0

data/lib/llm/providers/google/response_adapter/completion.rb

@@ -33,6 +33,36 @@ module LLM::Google::ResponseAdapter
       body.usageMetadata.thoughtsTokenCount || 0
     end
 
+    ##
+    # (see LLM::Contract::Completion#input_audio_tokens)
+    def input_audio_tokens
+      super
+    end
+
+    ##
+    # (see LLM::Contract::Completion#output_audio_tokens)
+    def output_audio_tokens
+      super
+    end
+
+    ##
+    # (see LLM::Contract::Completion#input_image_tokens)
+    def input_image_tokens
+      super
+    end
+
+    ##
+    # (see LLM::Contract::Completion#cache_read_tokens)
+    def cache_read_tokens
+      0
+    end
+
+    ##
+    # (see LLM::Contract::Completion#cache_write_tokens)
+    def cache_write_tokens
+      0
+    end
+
     ##
     # (see LLM::Contract::Completion#total_tokens)
     def total_tokens

data/lib/llm/providers/google.rb

@@ -208,7 +208,7 @@ module LLM
       req  = Net::HTTP::Post.new(path, headers)
       messages = build_complete_messages(prompt, params, role)
       body = LLM.json.dump({contents: adapt(messages)}.merge!(params))
-      set_body_stream(req, StringIO.new(body))
+      transport.set_body_stream(req, StringIO.new(body))
       req
     end
 

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

@@ -5,7 +5,7 @@ class LLM::Ollama
   # @private
   class ErrorHandler
     ##
-    # @return [Net::HTTPResponse]
+    # @return [LLM::Transport::Response]
     #  Non-2XX response from the server
     attr_reader :res
 
@@ -19,13 +19,13 @@ class LLM::Ollama
     #  The tracer
     # @param [Object, nil] span
     #  The span
-    # @param [Net::HTTPResponse] res
+    # @param [LLM::Transport::Response, Net::HTTPResponse] res
     #  The response from the server
     # @return [LLM::Ollama::ErrorHandler]
     def initialize(tracer, span, res)
       @tracer = tracer
       @span = span
-      @res = res
+      @res = LLM::Transport::Response.from(res)
     end
 
     ##
@@ -43,12 +43,11 @@ class LLM::Ollama
     ##
     # @return [LLM::Error]
     def error
-      case res
-      when Net::HTTPServerError
+      if res.server_error?
         LLM::ServerError.new("Server error").tap { _1.response = res }
-      when Net::HTTPUnauthorized
+      elsif res.unauthorized?
         LLM::UnauthorizedError.new("Authentication error").tap { _1.response = res }
-      when Net::HTTPTooManyRequests
+      elsif res.rate_limited?
         LLM::RateLimitError.new("Too many requests").tap { _1.response = res }
       else
         LLM::Error.new("Unexpected response").tap { _1.response = res }

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

@@ -17,8 +17,6 @@ class LLM::Ollama
   #     print "id: ", model.id, "\n"
   #   end
   class Models
-    include LLM::Utils
-
     ##
     # Returns a new Models object
     # @param provider [LLM::Provider]

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

@@ -27,6 +27,36 @@ module LLM::Ollama::ResponseAdapter
       0
     end
 
+    ##
+    # (see LLM::Contract::Completion#input_audio_tokens)
+    def input_audio_tokens
+      super
+    end
+
+    ##
+    # (see LLM::Contract::Completion#output_audio_tokens)
+    def output_audio_tokens
+      super
+    end
+
+    ##
+    # (see LLM::Contract::Completion#input_image_tokens)
+    def input_image_tokens
+      super
+    end
+
+    ##
+    # (see LLM::Contract::Completion#cache_read_tokens)
+    def cache_read_tokens
+      0
+    end
+
+    ##
+    # (see LLM::Contract::Completion#cache_write_tokens)
+    def cache_write_tokens
+      0
+    end
+
     ##
     # (see LLM::Contract::Completion#total_tokens)
     def total_tokens

data/lib/llm/providers/ollama.rb

@@ -130,7 +130,7 @@ module LLM
       messages = build_complete_messages(prompt, params, role)
       body = LLM.json.dump({messages: [adapt(messages)].flatten}.merge!(params))
       req = Net::HTTP::Post.new("/api/chat", headers)
-      set_body_stream(req, StringIO.new(body))
+      transport.set_body_stream(req, StringIO.new(body))
       req
     end
 

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

@@ -57,7 +57,7 @@ class LLM::OpenAI
       multi = LLM::Multipart.new(params.merge!(file: LLM.File(file), model:))
       req = Net::HTTP::Post.new(path("/audio/transcriptions"), headers)
       req["content-type"] = multi.content_type
-      set_body_stream(req, multi.body)
+      transport.set_body_stream(req, multi.body)
       res, span, tracer = execute(request: req, operation: "request")
       res = LLM::Response.new(res)
       tracer.on_request_finish(operation: "request", model:, res:, span:)
@@ -81,7 +81,7 @@ class LLM::OpenAI
       multi = LLM::Multipart.new(params.merge!(file: LLM.File(file), model:))
       req = Net::HTTP::Post.new(path("/audio/translations"), headers)
       req["content-type"] = multi.content_type
-      set_body_stream(req, multi.body)
+      transport.set_body_stream(req, multi.body)
       res, span, tracer = execute(request: req, operation: "request")
       res = LLM::Response.new(res)
       tracer.on_request_finish(operation: "request", model:, res:, span:)
@@ -90,7 +90,7 @@ class LLM::OpenAI
 
     private
 
-    [:path, :headers, :execute, :set_body_stream].each do |m|
+    [:path, :headers, :execute, :transport].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

@@ -5,7 +5,7 @@ class LLM::OpenAI
   # @private
   class ErrorHandler
     ##
-    # @return [Net::HTTPResponse]
+    # @return [LLM::Transport::Response]
     #  Non-2XX response from the server
     attr_reader :res
 
@@ -19,13 +19,13 @@ class LLM::OpenAI
     #  The tracer
     # @param [Object, nil] span
     #  The span
-    # @param [Net::HTTPResponse] res
+    # @param [LLM::Transport::Response, Net::HTTPResponse] res
     #  The response from the server
     # @return [LLM::OpenAI::ErrorHandler]
     def initialize(tracer, span, res)
       @tracer = tracer
       @span = span
-      @res = res
+      @res = LLM::Transport::Response.from(res)
     end
 
     ##
@@ -49,12 +49,11 @@ class LLM::OpenAI
     ##
     # @return [LLM::Error]
     def error
-      case res
-      when Net::HTTPServerError
+      if res.server_error?
         LLM::ServerError.new("Server error").tap { _1.response = res }
-      when Net::HTTPUnauthorized
+      elsif res.unauthorized?
         LLM::UnauthorizedError.new("Authentication error").tap { _1.response = res }
-      when Net::HTTPTooManyRequests
+      elsif res.rate_limited?
         LLM::RateLimitError.new("Too many requests").tap { _1.response = res }
       else
         error = body["error"] || {}

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

@@ -62,7 +62,7 @@ class LLM::OpenAI
       multi = LLM::Multipart.new(params.merge!(file: LLM.File(file), purpose:))
       req = Net::HTTP::Post.new(path("/files"), headers)
       req["content-type"] = multi.content_type
-      set_body_stream(req, multi.body)
+      transport.set_body_stream(req, multi.body)
       res, span, tracer = execute(request: req, operation: "request")
       res = ResponseAdapter.adapt(res, type: :file)
       tracer.on_request_finish(operation: "request", res:, span:)
@@ -134,7 +134,7 @@ class LLM::OpenAI
 
     private
 
-    [:path, :headers, :execute, :set_body_stream].each do |m|
+    [:path, :headers, :execute, :transport].each do |m|
       define_method(m) { |*args, **kwargs, &b| @provider.send(m, *args, **kwargs, &b) }
     end
   end

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

@@ -78,7 +78,7 @@ class LLM::OpenAI
       multi = LLM::Multipart.new(params.merge!(image:, model:, response_format:))
       req = Net::HTTP::Post.new(path("/images/variations"), headers)
       req["content-type"] = multi.content_type
-      set_body_stream(req, multi.body)
+      transport.set_body_stream(req, multi.body)
       res, span, tracer = execute(request: req, operation: "request")
       res = ResponseAdapter.adapt(res, type: :image)
       tracer.on_request_finish(operation: "request", model:, res:, span:)
@@ -104,7 +104,7 @@ class LLM::OpenAI
       multi = LLM::Multipart.new(params.merge!(image:, prompt:, model:, response_format:))
       req = Net::HTTP::Post.new(path("/images/edits"), headers)
       req["content-type"] = multi.content_type
-      set_body_stream(req, multi.body)
+      transport.set_body_stream(req, multi.body)
       res, span, tracer = execute(request: req, operation: "request")
       res = ResponseAdapter.adapt(res, type: :image)
       tracer.on_request_finish(operation: "request", model:, res:, span:)
@@ -113,7 +113,7 @@ class LLM::OpenAI
 
     private
 
-    [:path, :headers, :execute, :set_body_stream].each do |m|
+    [:path, :headers, :execute, :transport].each do |m|
       define_method(m) { |*args, **kwargs, &b| @provider.send(m, *args, **kwargs, &b) }
     end
   end

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

@@ -48,7 +48,7 @@ class LLM::OpenAI
 
     private
 
-    [:path, :headers, :execute, :set_body_stream].each do |m|
+    [:path, :headers, :execute].each do |m|
       define_method(m) { |*args, **kwargs, &b| @provider.send(m, *args, **kwargs, &b) }
     end
   end

data/lib/llm/providers/openai/response_adapter/completion.rb

@@ -40,6 +40,48 @@ module LLM::OpenAI::ResponseAdapter
         &.reasoning_tokens || 0
     end
 
+    ##
+    # (see LLM::Contract::Completion#input_audio_tokens)
+    def input_audio_tokens
+      body
+        .usage
+        &.prompt_tokens_details
+        &.audio_tokens || 0
+    end
+
+    ##
+    # (see LLM::Contract::Completion#output_audio_tokens)
+    def output_audio_tokens
+      body
+        .usage
+        &.completion_tokens_details
+        &.audio_tokens || 0
+    end
+
+    ##
+    # (see LLM::Contract::Completion#input_image_tokens)
+    def input_image_tokens
+      body
+        .usage
+        &.prompt_tokens_details
+        &.image_tokens || 0
+    end
+
+    ##
+    # (see LLM::Contract::Completion#cache_read_tokens)
+    def cache_read_tokens
+      body
+        .usage
+        &.prompt_tokens_details
+        &.cached_tokens || 0
+    end
+
+    ##
+    # (see LLM::Contract::Completion#cache_write_tokens)
+    def cache_write_tokens
+      0
+    end
+
     ##
     # (see LLM::Contract::Completion#total_tokens)
     def total_tokens

data/lib/llm/providers/openai/response_adapter/responds.rb

@@ -42,6 +42,45 @@ module LLM::OpenAI::ResponseAdapter
         &.reasoning_tokens || 0
     end
 
+    ##
+    # (see LLM::Contract::Completion#input_audio_tokens)
+    def input_audio_tokens
+      body
+        .usage
+        &.input_tokens_details
+        &.audio_tokens || 0
+    end
+
+    ##
+    # (see LLM::Contract::Completion#output_audio_tokens)
+    def output_audio_tokens
+      body
+        .usage
+        &.output_tokens_details
+        &.audio_tokens || 0
+    end
+
+    ##
+    # (see LLM::Contract::Completion#input_image_tokens)
+    def input_image_tokens
+      super
+    end
+
+    ##
+    # (see LLM::Contract::Completion#cache_read_tokens)
+    def cache_read_tokens
+      body
+        .usage
+        &.input_tokens_details
+        &.cached_tokens || 0
+    end
+
+    ##
+    # (see LLM::Contract::Completion#cache_write_tokens)
+    def cache_write_tokens
+      0
+    end
+
     ##
     # (see LLM::Contract::Completion#total_tokens)
     def total_tokens

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

@@ -44,7 +44,7 @@ class LLM::OpenAI
       messages = build_complete_messages(prompt, params, role)
       @provider.tracer.set_request_metadata(user_input: extract_user_input(messages, fallback: prompt))
       body = LLM.json.dump({input: [adapt(messages, mode: :response)].flatten}.merge!(params))
-      set_body_stream(req, StringIO.new(body))
+      transport.set_body_stream(req, StringIO.new(body))
       res, span, tracer = execute(request: req, stream:, stream_parser:, operation: "chat", model: params[:model])
       res = ResponseAdapter.adapt(res, type: :responds)
         .extend(Module.new { define_method(:__tools__) { tools } })
@@ -85,7 +85,7 @@ class LLM::OpenAI
 
     private
 
-    [:path, :headers, :execute, :set_body_stream, :resolve_tools].each do |m|
+    [:path, :headers, :execute, :transport, :resolve_tools].each do |m|
       define_method(m) { |*args, **kwargs, &b| @provider.send(m, *args, **kwargs, &b) }
     end
 

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

@@ -259,7 +259,7 @@ class LLM::OpenAI
 
     private
 
-    [:path, :headers, :execute, :set_body_stream].each do |m|
+    [:path, :headers, :execute].each do |m|
       define_method(m) { |*args, **kwargs, &b| @provider.send(m, *args, **kwargs, &b) }
     end
   end

data/lib/llm/providers/openai.rb

@@ -223,7 +223,7 @@ module LLM
       messages = build_complete_messages(prompt, params, role)
       body = LLM.json.dump({messages: adapt(messages, mode: :complete).flatten}.merge!(params))
       req = Net::HTTP::Post.new(completions_path, headers)
-      set_body_stream(req, StringIO.new(body))
+      transport.set_body_stream(req, StringIO.new(body))
       [req, messages]
     end
 

data/lib/llm/response.rb

@@ -10,25 +10,27 @@ module LLM
   # handling can share one common surface without flattening away
   # specialized behavior.
   #
-  # The normalized response still keeps the original
-  # {Net::HTTPResponse Net::HTTPResponse} available through {#res}
-  # when callers need direct access to raw HTTP details such as
-  # headers, status codes, or unadapted bodies.
+  # The normalized response keeps the transport response available
+  # through {#res}. When the default net/http transport is in use,
+  # {LLM::Transport::Response::HTTP
+  # LLM::Transport::Response::HTTP} keeps the
+  # original {Net::HTTPResponse Net::HTTPResponse} available through
+  # its own {LLM::Transport::Response::HTTP#res #res}.
   class Response
     require "json"
 
     ##
     # Returns the HTTP response
-    # @return [Net::HTTPResponse]
+    # @return [LLM::Transport::Response]
     attr_reader :res
 
     ##
-    # @param [Net::HTTPResponse] res
+    # @param [LLM::Transport::Response] res
     #  HTTP response
     # @return [LLM::Response]
     #  Returns an instance of LLM::Response
     def initialize(res)
-      @res = res
+      @res = LLM::Transport::Response.from(res)
     end
 
     ##
@@ -51,7 +53,7 @@ module LLM
     # Returns true if the response is successful
     # @return [Boolean]
     def ok?
-      Net::HTTPSuccess === @res
+      @res.success?
     end
 
     ##

data/lib/llm/sequel/plugin.rb

@@ -184,14 +184,6 @@ module LLM::Sequel
       ctx.wait(...)
     end
 
-    ##
-    # Calls into the stored context.
-    # @see LLM::Context#call
-    # @return [Object]
-    def call(...)
-      ctx.call(...)
-    end
-
     ##
     # @see LLM::Context#mode
     # @return [Symbol]
@@ -222,6 +214,13 @@ module LLM::Sequel
       ctx.functions
     end
 
+    ##
+    # @see LLM::Context#functions?
+    # @return [Boolean]
+    def functions?
+      ctx.functions?
+    end
+
     ##
     # @see LLM::Context#returns
     # @return [Array<LLM::Function::Return>]

data/lib/llm/stream/queue.rb

@@ -4,7 +4,7 @@ class LLM::Stream
   ##
   # A small queue for collecting streamed tool work. Values can be immediate
   # {LLM::Function::Return} objects or concurrent handles returned by
-  # {LLM::Function#spawn}. Calling {#wait(strategy)} resolves queued work and
+  # {LLM::Function#spawn}. Calling {#wait} resolves queued work and
   # returns an array of {LLM::Function::Return} values.
   class Queue
     ##
@@ -41,56 +41,29 @@ class LLM::Stream
 
     ##
     # Waits for queued work to finish and returns function results.
-    # @param [Symbol, Array<Symbol>] strategy
-    #   Controls concurrency strategy, or lists the possible concurrency strategies
-    #   to wait on:
-    #   - `:thread`: Use threads
-    #   - `:task`: Use async tasks (requires async gem)
-    #   - `:fiber`: Use scheduler-backed fibers (requires Fiber.scheduler)
-    #   - `:ractor`: Use Ruby ractors (class-based tools only; MCP tools are not supported)
-    #   - `[:thread, :ractor]`: Wait for any queued thread or ractor work, in the
-    #     given order. This is useful when different tools were spawned with
-    #     different concurrency strategies.
+    #
+    # Queued work is waited according to the actual task types that were
+    # enqueued, so callers do not need to provide a strategy here.
+    #
     # @return [Array<LLM::Function::Return>]
-    def wait(strategy)
+    def wait
       returns, tasks = @items.shift(@items.length).partition { LLM::Function::Return === _1 }
-      results = wait_tasks(tasks, strategy)
+      results = wait_tasks(tasks)
       returns.concat fire_hooks(tasks, results)
     end
     alias_method :value, :wait
 
     private
 
-    def wait_tasks(tasks, strategy)
-      strategies = Array(strategy)
-      return wait_group(tasks, strategies.first) unless strategies.length > 1
-      grouped = strategies.to_h { [_1, []] }
-      tasks.each do |task|
-        grouped[task_strategy(task)] << task
-      end
-      strategies.flat_map do |name|
-        selected = grouped.fetch(name)
-        selected.empty? ? [] : wait_group(selected, name)
-      end
-    end
-
-    def wait_group(tasks, strategy)
-      case strategy
-      when :thread then LLM::Function::ThreadGroup.new(tasks).wait
-      when :task then LLM::Function::TaskGroup.new(tasks).wait
-      when :fiber then LLM::Function::FiberGroup.new(tasks).wait
-      when :ractor then LLM::Function::Ractor::Group.new(tasks).wait
-      else raise ArgumentError, "Unknown strategy: #{strategy.inspect}. Expected :thread, :task, :fiber, or :ractor"
-      end
-    end
-
-    def task_strategy(task)
-      case task.task
-      when Thread then :thread
-      when Fiber then :fiber
-      when LLM::Function::Ractor::Task then :ractor
-      else :task
+    def wait_tasks(tasks)
+      return [] if tasks.empty?
+      results = {}
+      grouped_tasks = tasks.group_by(&:group_class)
+      grouped_tasks.each do |group_class, group|
+        returns = group_class.new(group).wait
+        returns.each.with_index { results[group[_2]] = _1 }
       end
+      tasks.map { results[_1] }
     end
 
     def fire_hooks(tasks, results)

data/lib/llm/stream.rb

@@ -46,11 +46,11 @@ module LLM
 
     ##
     # Waits for queued tool work to finish and returns function results.
-    # @param [Symbol] strategy
-    #  The concurrency strategy to use
+    # Any passed arguments are ignored because queued work is waited according
+    # to the actual task types already present in the queue.
     # @return [Array<LLM::Function::Return>]
-    def wait(strategy)
-      queue.wait(strategy)
+    def wait(*)
+      queue.wait
     end
 
     # @group Public callbacks

data/lib/llm/transport/execution.rb

@@ -0,0 +1,67 @@
+# frozen_string_literal: true
+
+class LLM::Transport
+  ##
+  # Internal request execution methods for {LLM::Provider}.
+  #
+  # This module handles provider-side transport execution, response
+  # parsing, streaming, and request body setup.
+  #
+  # @api private
+  module Execution
+    private
+
+    ##
+    # Executes a HTTP request
+    # @param [Net::HTTPRequest] request
+    #  The request to send
+    # @param [Proc] b
+    #  A block to yield the response to (optional)
+    # @return [LLM::Transport::Response]
+    #  The response from the server
+    # @raise [LLM::Error::Unauthorized]
+    #  When authentication fails
+    # @raise [LLM::Error::RateLimit]
+    #  When the rate limit is exceeded
+    # @raise [LLM::Error]
+    #  When any other unsuccessful status code is returned
+    # @raise [SystemCallError]
+    #  When there is a network error at the operating system level
+    # @return [LLM::Transport::Response]
+    def execute(request:, operation:, stream: nil, stream_parser: self.stream_parser, model: nil, inputs: nil, &b)
+      stream &&= LLM::Object.from(streamer: stream, parser: stream_parser, decoder: stream_decoder)
+      owner = transport.request_owner
+      tracer = self.tracer
+      span = tracer.on_request_start(operation:, model:, inputs:)
+      res = transport.request(request, owner:, stream:, &b)
+      res = LLM::Transport::Response.from(res)
+      [handle_response(res, tracer, span), span, tracer]
+    rescue *transport.interrupt_errors
+      raise LLM::Interrupt, "request interrupted" if transport.interrupted?(owner)
+      raise
+    end
+
+    ##
+    # Handles the response from a request
+    # @param [LLM::Transport::Response] res
+    #  The response to handle
+    # @param [Object, nil] span
+    #  The span
+    # @return [LLM::Transport::Response]
+    def handle_response(res, tracer, span)
+      res.ok? ? res.body = parse_response(res) : error_handler.new(tracer, span, res).raise_error!
+      res
+    end
+
+    ##
+    # Parse a HTTP response
+    # @param [LLM::Transport::Response] res
+    # @return [LLM::Object, String]
+    def parse_response(res)
+      case res["content-type"]
+      when %r{\Aapplication/json\s*} then LLM::Object.from(LLM.json.load(res.body))
+      else res.body
+      end
+    end
+  end
+end