llm.rb 0.3.2 → 0.4.0

data/lib/llm/providers/openai.rb

@@ -12,6 +12,7 @@ module LLM
     require_relative "openai/images"
     require_relative "openai/audio"
     require_relative "openai/files"
+    require_relative "openai/models"
     include Format
 
     HOST = "api.openai.com"
@@ -43,16 +44,23 @@ module LLM
     # @param prompt (see LLM::Provider#complete)
     # @param role (see LLM::Provider#complete)
     # @param model (see LLM::Provider#complete)
+    # @param schema (see LLM::Provider#complete)
     # @param params (see LLM::Provider#complete)
     # @example (see LLM::Provider#complete)
     # @raise (see LLM::Provider#request)
+    # @raise [LLM::Error::PromptError]
+    #  When given an object a provider does not understand
     # @return (see LLM::Provider#complete)
-    def complete(prompt, role = :user, model: "gpt-4o-mini", **params)
-      params   = {model:}.merge!(params)
-      req      = Net::HTTP::Post.new("/v1/chat/completions", headers)
+    def complete(prompt, role = :user, model: default_model, schema: nil, **params)
+      params = {model:}
+                 .merge!(expand_schema(schema))
+                 .merge!(params)
+                 .compact
+      req = Net::HTTP::Post.new("/v1/chat/completions", headers)
       messages = [*(params.delete(:messages) || []), Message.new(role, prompt)]
-      req.body = JSON.dump({messages: format(messages, :complete)}.merge!(params))
-      res      = request(@http, req)
+      body = JSON.dump({messages: format(messages, :complete)}.merge!(params))
+      set_body_stream(req, StringIO.new(body))
+      res = request(@http, req)
       Response::Completion.new(res).extend(response_parser)
     end
 
@@ -88,14 +96,26 @@ module LLM
       LLM::OpenAI::Files.new(self)
     end
 
+    ##
+    # Provides an interface to OpenAI's models API
+    # @see https://platform.openai.com/docs/api-reference/models/list OpenAI docs
+    # @return [LLM::OpenAI::Models]
+    def models
+      LLM::OpenAI::Models.new(self)
+    end
+
     ##
     # @return (see LLM::Provider#assistant_role)
     def assistant_role
       "assistant"
     end
 
-    def models
-      @models ||= load_models!("openai")
+    ##
+    # Returns the default model for chat completions
+    # @see https://platform.openai.com/docs/models/gpt-4o-mini gpt-4o-mini
+    # @return [String]
+    def default_model
+      "gpt-4o-mini"
     end
 
     private
@@ -114,5 +134,15 @@ module LLM
     def error_handler
       LLM::OpenAI::ErrorHandler
     end
+
+    def expand_schema(schema)
+      return {} unless schema
+      {
+        response_format: {
+          type: "json_schema",
+          json_schema: {name: "JSONSchema", schema:}
+        }
+      }
+    end
   end
 end

data/lib/llm/response/modellist.rb

@@ -0,0 +1,18 @@
+# frozen_string_literal: true
+
+module LLM
+  ##
+  # The {LLM::Response::ModelList LLM::Response::ModelList} class represents a
+  # list of model objects that are returned by a provider. It is an Enumerable
+  # object, and can be used to iterate over the model objects in a way that is
+  # similar to an array. Each element is an instance of OpenStruct.
+  class Response::ModelList < Response
+    include Enumerable
+
+    attr_accessor :models
+
+    def each(&)
+      @models.each(&)
+    end
+  end
+end

data/lib/llm/response.rb

@@ -13,6 +13,7 @@ module LLM
     require_relative "response/file"
     require_relative "response/filelist"
     require_relative "response/download_file"
+    require_relative "response/modellist"
 
     ##
     # @param [Net::HTTPResponse] res

data/lib/llm/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module LLM
-  VERSION = "0.3.2"
+  VERSION = "0.4.0"
 end

data/lib/llm.rb

@@ -1,6 +1,8 @@
 # frozen_string_literal: true
 
 module LLM
+  require "stringio"
+  require_relative "llm/core_ext/ostruct"
   require_relative "llm/version"
   require_relative "llm/utils"
   require_relative "llm/error"
@@ -13,7 +15,6 @@ module LLM
   require_relative "llm/provider"
   require_relative "llm/chat"
   require_relative "llm/buffer"
-  require_relative "llm/core_ext/ostruct"
 
   module_function
 

data/spec/anthropic/completion_spec.rb

@@ -42,6 +42,42 @@ RSpec.describe "LLM::Anthropic: completions" do
     end
   end
 
+  context "when given a URI to an image",
+          vcr: {cassette_name: "anthropic/completions/successful_response_uri_image"} do
+    subject { response.choices[0].content.downcase[0..2] }
+    let(:response) do
+      anthropic.complete([
+        "Is this image the flag of brazil ? ",
+        "Answer with yes or no. ",
+        "Nothing else.",
+        uri
+      ], :user)
+    end
+    let(:uri) { URI("https://upload.wikimedia.org/wikipedia/en/thumb/0/05/Flag_of_Brazil.svg/250px-Flag_of_Brazil.svg.png") }
+
+    it "describes the image" do
+      is_expected.to eq("yes")
+    end
+  end
+
+  context "when given a local reference to an image",
+          vcr: {cassette_name: "anthropic/completions/successful_response_file_image"} do
+    subject { response.choices[0].content.downcase[0..2] }
+    let(:response) do
+      anthropic.complete([
+        "Is this image a representation of a blue book ?",
+        "Answer with yes or no.",
+        "Nothing else.",
+        file
+      ], :user)
+    end
+    let(:file) { LLM::File("spec/fixtures/images/bluebook.png") }
+
+    it "describes the image" do
+      is_expected.to eq("yes")
+    end
+  end
+
   context "when given an unauthorized response",
           vcr: {cassette_name: "anthropic/completions/unauthorized_response"} do
     subject(:response) { anthropic.complete("Hello", :user) }

data/spec/anthropic/models_spec.rb

@@ -0,0 +1,21 @@
+# frozen_string_literal: true
+
+require "setup"
+
+RSpec.describe "LLM::Anthropic::Models" do
+  let(:token) { ENV["LLM_SECRET"] || "TOKEN" }
+  let(:provider) { LLM.anthropic(token) }
+
+  context "when given a successful list operation",
+          vcr: {cassette_name: "anthropic/models/successful_list"} do
+    subject { provider.models.all }
+
+    it "is successful" do
+      is_expected.to be_instance_of(LLM::Response::ModelList)
+    end
+
+    it "returns a list of models" do
+      expect(subject.models).to all(be_a(LLM::Model))
+    end
+  end
+end

data/spec/gemini/images_spec.rb

@@ -14,12 +14,8 @@ RSpec.describe "LLM::Gemini::Images" do
       expect(response).to be_instance_of(LLM::Response::Image)
     end
 
-    it "returns an encoded string" do
-      expect(response.images[0].encoded).to be_instance_of(String)
-    end
-
-    it "returns a binary string" do
-      expect(response.images[0].binary).to be_instance_of(String)
+    it "returns an IO-like object" do
+      expect(response.images[0]).to be_instance_of(StringIO)
     end
   end
 
@@ -36,12 +32,8 @@ RSpec.describe "LLM::Gemini::Images" do
       expect(response).to be_instance_of(LLM::Response::Image)
     end
 
-    it "returns data" do
-      expect(response.images[0].encoded).to be_instance_of(String)
-    end
-
-    it "returns a url" do
-      expect(response.images[0].binary).to be_instance_of(String)
+    it "returns an IO-like object" do
+      expect(response.images[0]).to be_instance_of(StringIO)
     end
   end
 end

data/spec/gemini/models_spec.rb

@@ -0,0 +1,21 @@
+# frozen_string_literal: true
+
+require "setup"
+
+RSpec.describe "LLM::Gemini::Models" do
+  let(:token) { ENV["LLM_SECRET"] || "TOKEN" }
+  let(:provider) { LLM.gemini(token) }
+
+  context "when given a successful list operation",
+          vcr: {cassette_name: "gemini/models/successful_list"} do
+    subject { provider.models.all }
+
+    it "is successful" do
+      is_expected.to be_instance_of(LLM::Response::ModelList)
+    end
+
+    it "returns a list of models" do
+      expect(subject.models).to all(be_a(LLM::Model))
+    end
+  end
+end

data/spec/llm/conversation_spec.rb

@@ -27,7 +27,7 @@ RSpec.describe "LLM::Chat: non-lazy" do
         bot = nil
         inputs.zip(outputs).each_with_index do |(input, output), index|
           expect(provider).to receive(:complete)
-                                .with(input.content, instance_of(Symbol), messages:)
+                                .with(input.content, instance_of(Symbol), messages:, model: provider.default_model, schema: nil)
                                 .and_return(OpenStruct.new(choices: [output]))
           bot = index.zero? ? provider.chat!(input.content, :system) : bot.chat(input.content)
           messages.concat([input, output])
@@ -117,7 +117,8 @@ RSpec.describe "LLM::Chat: lazy" do
 
       context "when given a specific model",
               vcr: {cassette_name: "openai/lazy_conversation/completions/successful_response_o3_mini"} do
-        let(:conversation) { described_class.new(provider, model: provider.models["o3-mini"]).lazy }
+        let(:model) { provider.models.all.find { _1.id == "o3-mini" } }
+        let(:conversation) { described_class.new(provider, model:).lazy }
 
         it "maintains the model throughout a conversation" do
           conversation.chat(prompt, :system)
@@ -179,7 +180,8 @@ RSpec.describe "LLM::Chat: lazy" do
 
       context "when given a specific model",
               vcr: {cassette_name: "openai/lazy_conversation/responses/successful_response_o3_mini"} do
-        let(:conversation) { described_class.new(provider, model: provider.models["o3-mini"]).lazy }
+        let(:model) { provider.models.all.find { _1.id == "o3-mini" } }
+        let(:conversation) { described_class.new(provider, model:).lazy }
 
         it "maintains the model throughout a conversation" do
           conversation.respond(prompt, :developer)
@@ -190,4 +192,70 @@ RSpec.describe "LLM::Chat: lazy" do
       end
     end
   end
+
+  context "when given a schema as JSON" do
+    context "with openai" do
+      let(:provider) { LLM.openai(token) }
+      let(:conversation) { described_class.new(provider, schema:).lazy }
+
+      context "when given a schema",
+              vcr: {cassette_name: "openai/lazy_conversation/completions/successful_response_schema_netbsd"} do
+        subject(:message) { conversation.recent_message.content! }
+        let(:schema) { provider.schema.object({os: provider.schema.string.enum("OpenBSD", "FreeBSD", "NetBSD")}) }
+
+        before do
+          conversation.chat "You secretly love NetBSD", :system
+          conversation.chat "What operating system is the best?", :user
+        end
+
+        it "formats the response" do
+          is_expected.to eq("os" => "NetBSD")
+        end
+      end
+    end
+
+    context "with gemini" do
+      let(:provider) { LLM.gemini(token) }
+      let(:conversation) { described_class.new(provider, schema:).lazy }
+
+      context "when given a schema",
+              vcr: {cassette_name: "gemini/lazy_conversation/completions/successful_response_schema_netbsd"} do
+        subject(:message) { conversation.recent_message.content! }
+        let(:schema) { provider.schema.object({os: provider.schema.string.enum("OpenBSD", "FreeBSD", "NetBSD")}) }
+
+        before do
+          conversation.chat "You secretly love NetBSD", :user
+          conversation.chat "What operating system is the best?", :user
+        end
+
+        it "formats the response" do
+          is_expected.to eq("os" => "NetBSD")
+        end
+      end
+    end
+
+    context "with ollama" do
+      let(:provider) { LLM.ollama(nil, host: "eel.home.network") }
+      let(:conversation) { described_class.new(provider, schema:).lazy }
+
+      context "when given a schema",
+              vcr: {cassette_name: "ollama/lazy_conversation/completions/successful_response_schema_netbsd"} do
+        subject(:message) { conversation.recent_message.content! }
+        let(:schema) do
+          provider.schema.object({
+            os: provider.schema.string.enum("OpenBSD", "FreeBSD", "NetBSD").required
+          })
+        end
+
+        before do
+          conversation.chat "You secretly love NetBSD", :system
+          conversation.chat "What operating system is the best?", :user
+        end
+
+        it "formats the response" do
+          is_expected.to eq("os" => "NetBSD")
+        end
+      end
+    end
+  end
 end

data/spec/ollama/models_spec.rb

@@ -0,0 +1,20 @@
+# frozen_string_literal: true
+
+require "setup"
+
+RSpec.describe "LLM::Ollama::Models" do
+  let(:provider) { LLM.ollama(nil, host: "eel.home.network") }
+
+  context "when given a successful list operation",
+          vcr: {cassette_name: "ollama/models/successful_list"} do
+    subject { provider.models.all }
+
+    it "is successful" do
+      is_expected.to be_instance_of(LLM::Response::ModelList)
+    end
+
+    it "returns a list of models" do
+      expect(subject.models).to all(be_a(LLM::Model))
+    end
+  end
+end

data/spec/openai/completion_spec.rb

@@ -64,6 +64,25 @@ RSpec.describe "LLM::OpenAI: completions" do
     end
   end
 
+  context "when asked to describe an audio file",
+          vcr: {cassette_name: "openai/completions/describe_pdf_document"} do
+    let(:file) { LLM::File("spec/fixtures/documents/freebsd.sysctl.pdf") }
+    let(:response) do
+      openai.complete([
+        "This PDF document describes sysctl nodes on FreeBSD",
+        "Answer yes or no.",
+        "Nothing else",
+        file
+      ], :user)
+    end
+
+    subject { response.choices[0].content.downcase[0..2] }
+
+    it "is successful" do
+      is_expected.to eq("yes")
+    end
+  end
+
   context "when given a 'bad request' response",
           vcr: {cassette_name: "openai/completions/bad_request"} do
     subject(:response) { openai.complete(URI("/foobar.exe"), :user) }

data/spec/openai/images_spec.rb

@@ -40,12 +40,8 @@ RSpec.describe "LLM::OpenAI::Images" do
       expect(response.images).to be_instance_of(Array)
     end
 
-    it "returns an encoded string" do
-      expect(response.images[0].encoded).to be_instance_of(String)
-    end
-
-    it "returns an binary string" do
-      expect(response.images[0].binary).to be_instance_of(String)
+    it "returns an IO-like object" do
+      expect(response.images[0]).to be_instance_of(StringIO)
     end
   end
 

data/spec/openai/models_spec.rb

@@ -0,0 +1,21 @@
+# frozen_string_literal: true
+
+require "setup"
+
+RSpec.describe "LLM::OpenAI::Models" do
+  let(:token) { ENV["LLM_SECRET"] || "TOKEN" }
+  let(:provider) { LLM.openai(token) }
+
+  context "when given a successful list operation",
+          vcr: {cassette_name: "openai/models/successful_list"} do
+    subject { provider.models.all }
+
+    it "is successful" do
+      is_expected.to be_instance_of(LLM::Response::ModelList)
+    end
+
+    it "returns a list of models" do
+      expect(subject.models).to all(be_a(LLM::Model))
+    end
+  end
+end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llm.rb
 version: !ruby/object:Gem::Version
-  version: 0.3.2
+  version: 0.4.0
 platform: ruby
 authors:
 - Antar Azri
@@ -9,7 +9,7 @@ authors:
 autorequire:
 bindir: bin
 cert_chain: []
-date: 2025-04-26 00:00:00.000000000 Z
+date: 2025-04-29 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: webmock
@@ -148,6 +148,15 @@ extensions: []
 extra_rdoc_files: []
 files:
 - README.md
+- lib/json/schema.rb
+- lib/json/schema/array.rb
+- lib/json/schema/boolean.rb
+- lib/json/schema/integer.rb
+- lib/json/schema/leaf.rb
+- lib/json/schema/null.rb
+- lib/json/schema/number.rb
+- lib/json/schema/object.rb
+- lib/json/schema/string.rb
 - lib/llm.rb
 - lib/llm/buffer.rb
 - lib/llm/chat.rb
@@ -162,6 +171,7 @@ files:
 - lib/llm/providers/anthropic.rb
 - lib/llm/providers/anthropic/error_handler.rb
 - lib/llm/providers/anthropic/format.rb
+- lib/llm/providers/anthropic/models.rb
 - lib/llm/providers/anthropic/response_parser.rb
 - lib/llm/providers/gemini.rb
 - lib/llm/providers/gemini/audio.rb
@@ -169,10 +179,12 @@ files:
 - lib/llm/providers/gemini/files.rb
 - lib/llm/providers/gemini/format.rb
 - lib/llm/providers/gemini/images.rb
+- lib/llm/providers/gemini/models.rb
 - lib/llm/providers/gemini/response_parser.rb
 - lib/llm/providers/ollama.rb
 - lib/llm/providers/ollama/error_handler.rb
 - lib/llm/providers/ollama/format.rb
+- lib/llm/providers/ollama/models.rb
 - lib/llm/providers/ollama/response_parser.rb
 - lib/llm/providers/openai.rb
 - lib/llm/providers/openai/audio.rb
@@ -180,6 +192,7 @@ files:
 - lib/llm/providers/openai/files.rb
 - lib/llm/providers/openai/format.rb
 - lib/llm/providers/openai/images.rb
+- lib/llm/providers/openai/models.rb
 - lib/llm/providers/openai/response_parser.rb
 - lib/llm/providers/openai/responses.rb
 - lib/llm/providers/voyageai.rb
@@ -195,30 +208,31 @@ files:
 - lib/llm/response/file.rb
 - lib/llm/response/filelist.rb
 - lib/llm/response/image.rb
+- lib/llm/response/modellist.rb
 - lib/llm/response/output.rb
 - lib/llm/utils.rb
 - lib/llm/version.rb
 - llm.gemspec
-- share/llm/models/anthropic.yml
-- share/llm/models/gemini.yml
-- share/llm/models/ollama.yml
-- share/llm/models/openai.yml
 - spec/anthropic/completion_spec.rb
 - spec/anthropic/embedding_spec.rb
+- spec/anthropic/models_spec.rb
 - spec/gemini/completion_spec.rb
 - spec/gemini/conversation_spec.rb
 - spec/gemini/embedding_spec.rb
 - spec/gemini/files_spec.rb
 - spec/gemini/images_spec.rb
+- spec/gemini/models_spec.rb
 - spec/llm/conversation_spec.rb
 - spec/ollama/completion_spec.rb
 - spec/ollama/conversation_spec.rb
 - spec/ollama/embedding_spec.rb
+- spec/ollama/models_spec.rb
 - spec/openai/audio_spec.rb
 - spec/openai/completion_spec.rb
 - spec/openai/embedding_spec.rb
 - spec/openai/files_spec.rb
 - spec/openai/images_spec.rb
+- spec/openai/models_spec.rb
 - spec/openai/responses_spec.rb
 - spec/readme_spec.rb
 - spec/setup.rb

data/share/llm/models/anthropic.yml

@@ -1,35 +0,0 @@
-claude-3-7-sonnet-20250219:
-  name: Claude 3.7 Sonnet
-  parameters: Unknown
-  description: Most intelligent Claude model with extended thinking and high capability
-  to_param: claude-3-7-sonnet-20250219
-
-claude-3-5-sonnet-20241022:
-  name: Claude 3.5 Sonnet (v2)
-  parameters: Unknown
-  description: High intelligence and capability; upgraded from previous Sonnet
-  to_param: claude-3-5-sonnet-20241022
-
-claude-3-5-sonnet-20240620:
-  name: Claude 3.5 Sonnet
-  parameters: Unknown
-  description: Intelligent and capable general-purpose model
-  to_param: claude-3-5-sonnet-20240620
-
-claude-3-5-haiku-20241022:
-  name: Claude 3.5 Haiku
-  parameters: Unknown
-  description: Blazing fast model for low-latency text generation
-  to_param: claude-3-5-haiku-20241022
-
-claude-3-opus-20240229:
-  name: Claude 3 Opus
-  parameters: Unknown
-  description: Top-level intelligence, fluency, and reasoning for complex tasks
-  to_param: claude-3-opus-20240229
-
-claude-3-haiku-20240307:
-  name: Claude 3 Haiku
-  parameters: Unknown
-  description: Fastest and most compact Claude model for near-instant responsiveness
-  to_param: claude-3-haiku-20240307

data/share/llm/models/gemini.yml

@@ -1,35 +0,0 @@
-gemini-2.5-pro-exp-03-25:
-  name: Gemini
-  parameters: Unknown
-  description: Enhanced thinking and reasoning, multimodal understanding, advanced coding, and more
-  to_param: gemini-2.5-pro-exp-03-25
-
-gemini-2.0-flash:
-  name: Gemini
-  parameters: Unknown
-  description: Next generation features, speed, thinking, realtime streaming, and multimodal generation
-  to_param: gemini-2.0-flash
-
-gemini-2.0-flash-lite:
-  name: Gemini
-  parameters: Unknown
-  description: Cost efficiency and low latency
-  to_param: gemini-2.0-flash-lite
-
-gemini-1.5-flash:
-  name: Gemini
-  parameters: Unknown
-  description: Fast and versatile performance across a diverse variety of tasks
-  to_param: gemini-1.5-flash
-
-gemini-1.5-flash-8b:
-  name: Gemini
-  parameters: 8B
-  description: High volume and lower intelligence tasks
-  to_param: gemini-1.5-flash-8b
-
-gemini-1.5-pro:
-  name: Gemini
-  parameters: Unknown
-  description: Complex reasoning tasks requiring more intelligence
-  to_param: gemini-1.5-pro