nitro_intelligence 2.0.0 → 2.1.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 9c03a8a4c436205c01d926c024c1bcd12542553f36e3998b10462653cec4cdc1
-  data.tar.gz: 58f340818e47d7b6e36dd84f44b4072bf8636a83d4fd6931e65e62ec1ed29ab5
+  metadata.gz: 7d1c0e6c9ddffc032a989ebffb001d58aefabf86e50f92966a6d384a5218628f
+  data.tar.gz: 60b9ae5ef4fd8c7a79bc45f13a1b965c9d9a0792f38bc285e361822b7d91a8ad
 SHA512:
-  metadata.gz: 70eab830245ab3c7bd4a6835c080770ee324dca9c440ba200f0ac29fc528e9b4be5579091259a9abed7c35ce51372da82d4452d1df71ee26c51a61bdb342acaa
-  data.tar.gz: 549cd0d2477d1382cff28d866266797983b0ddd860396c10ff5edd8087f5538a4834b1194d1ae6d5d16d497825bf660308bcacfcb3da12846e2f69ef611264a4
+  metadata.gz: a13a9b1a0a58f58c142ab87ec89f491f409341717b3101b9bb1579c88b8ed9335413ea73d621c489749b7d151d62672e9277e3d39da80c4721008cad28a301b0
+  data.tar.gz: 1c498a0a4ec99af32f039dd43272ad61650da433c4f4d3b4297aadd473fba4c93af8d9ed08db0d2faf5012ed3a721e6d47587bf5429eb16368161d224f773cf7

data/lib/nitro_intelligence/agent_server.rb

@@ -10,6 +10,9 @@ module NitroIntelligence
     class RunError < StandardError; end
     class ThreadResumptionError < StandardError; end
 
+    # Aegra answers with a conflict when `ifExists: "raise"` is sent for a thread that already exists.
+    THREAD_CONFLICT_CODE = 409
+
     attr_reader :base_url, :user_id
 
     def initialize(base_url:, api_key:, user_id: "default-user")
@@ -21,6 +24,7 @@ module NitroIntelligence
       @api_key = api_key
       @user_id = user_id
       @tool_call_review_validator = ToolCallReviewValidator.new
+      @graph_ids = {}
     end
 
     def await_run(thread_id:, assistant_id:, messages:, context: {})
@@ -29,7 +33,7 @@ module NitroIntelligence
       initial_state = messages[0..-2]
       last_message = messages.last
 
-      initialize_thread_if_needed(thread_id:, initial_state:)
+      initialize_thread_if_needed(thread_id:, assistant_id:, initial_state:)
       trigger_run(thread_id:, assistant_id:, context:, last_message:)
     end
 
@@ -77,20 +81,78 @@ module NitroIntelligence
 
   private
 
-    def initialize_thread_if_needed(thread_id:, initial_state:)
-      thread_response = post(
+    # Aegra accepts `initial_state` on thread creation but never applies it, so a brand new thread is
+    # seeded through the thread state endpoint instead. A thread that already exists is left untouched:
+    # its state was seeded when it was created and has been built up by every run since.
+    def initialize_thread_if_needed(thread_id:, assistant_id:, initial_state:)
+      thread_response = create_thread(thread_id:, assistant_id:)
+
+      return if thread_already_exists?(thread_response)
+      raise ThreadInitializationError, thread_response.body if thread_response.code.to_i != 200
+      return if initial_state.blank?
+
+      seed_new_thread_state(thread_id:, initial_state:)
+    end
+
+    # Creating the thread and seeding its state are separate requests, so a failure between them leaves
+    # an empty thread behind. A retry would find that thread, take it for one already under way, skip
+    # seeding and run without the history -- losing the very thing seeding exists for, without an error.
+    # Discard the thread instead, so a retry starts over from a clean slate.
+    def seed_new_thread_state(thread_id:, initial_state:)
+      seed_thread_state(thread_id:, initial_state:)
+    rescue
+      discard_thread(thread_id:)
+      raise
+    end
+
+    def discard_thread(thread_id:)
+      delete(path: "/threads/#{thread_id}")
+    rescue
+      # Best effort. The seeding failure is the one worth surfacing, and it is raised either way.
+    end
+
+    def create_thread(thread_id:, assistant_id:)
+      post(
         path: "/threads",
         body: {
           threadId: thread_id.to_s,
-          ifExists: "do_nothing",
-          initial_state: { messages: initial_state },
+          ifExists: "raise",
+          # Without a graph_id, the thread state cannot be updated before the thread's first run.
+          metadata: { graph_id: graph_id_for(assistant_id) },
           user_id:,
         }
       )
+    end
 
-      raise ThreadInitializationError, thread_response.body if thread_response.code.to_i != 200
+    def graph_id_for(assistant_id)
+      @graph_ids[assistant_id] ||= fetch_graph_id(assistant_id)
+    end
 
-      JSON.parse(thread_response.body)
+    def fetch_graph_id(assistant_id)
+      assistant_response = get(path: "/assistants/#{assistant_id}")
+
+      raise ThreadInitializationError, assistant_response.body if assistant_response.code.to_i != 200
+
+      graph_id = JSON.parse(assistant_response.body)["graph_id"]
+
+      raise ThreadInitializationError, "Assistant #{assistant_id} has no graph_id" if graph_id.blank?
+
+      graph_id
+    end
+
+    def seed_thread_state(thread_id:, initial_state:)
+      state_response = post(
+        path: "/threads/#{thread_id}/state",
+        body: { values: { messages: initial_state } }
+      )
+
+      raise ThreadInitializationError, state_response.body if state_response.code.to_i != 200
+
+      JSON.parse(state_response.body)
+    end
+
+    def thread_already_exists?(response)
+      response.code.to_i == THREAD_CONFLICT_CODE
     end
 
     def get_thread_state(thread_id:)
@@ -177,6 +239,17 @@ module NitroIntelligence
       http.request(request)
     end
 
+    def delete(path:)
+      uri = URI("#{base_url}#{path}")
+      http = Net::HTTP.new(uri.host, uri.port)
+      http.use_ssl = uri.scheme == "https"
+
+      request = Net::HTTP::Delete.new(uri)
+      request_headers.each { |k, v| request[k] = v }
+
+      http.request(request)
+    end
+
     def post(path:, body:)
       uri = URI("#{base_url}#{path}")
       http = Net::HTTP.new(uri.host, uri.port)

data/lib/nitro_intelligence/client/handlers/audio_transcription_handler.rb

@@ -1,21 +1,19 @@
 require "openai"
+require "nitro_intelligence/client/handlers/base_handler"
 
 module NitroIntelligence
   module Client
     module Handlers
-      class AudioTranscriptionHandler
+      class AudioTranscriptionHandler < BaseHandler
         ALLOWED_EXTRA_PARAMETERS = OpenAI::Models::Audio::TranscriptionCreateParams.fields.keys.uniq.freeze
 
-        def initialize(client:)
-          @client = client
-        end
-
         def create(audio_file:, message: "", parameters: {})
           validate_and_resolve!(parameters)
           perform_request(audio_file:, message:, parameters:)
         end
 
         def perform_request(audio_file:, message: "", parameters: {})
+          add_request_headers(parameters, MODALITY_HEADER => "audio", REQUESTED_MODEL_HEADER => parameters[:model])
           @client.audio.transcriptions.create(
             prompt: message,
             file: audio_file,

data/lib/nitro_intelligence/client/handlers/base_handler.rb

@@ -0,0 +1,22 @@
+module NitroIntelligence
+  module Client
+    module Handlers
+      class BaseHandler
+        MODALITY_HEADER = "nip-modality".freeze
+        REQUESTED_MODEL_HEADER = "nip-requested-model".freeze
+
+        def initialize(client:)
+          @client = client
+        end
+
+      private
+
+        def add_request_headers(parameters, headers)
+          request_options = (parameters[:request_options] ||= {})
+          (request_options[:extra_headers] ||= {}).merge!(headers.compact)
+          parameters
+        end
+      end
+    end
+  end
+end

data/lib/nitro_intelligence/client/handlers/chat_handler.rb

@@ -1,21 +1,19 @@
 require "openai"
+require "nitro_intelligence/client/handlers/base_handler"
 
 module NitroIntelligence
   module Client
     module Handlers
-      class ChatHandler
+      class ChatHandler < BaseHandler
         ALLOWED_EXTRA_PARAMETERS = OpenAI::Models::Chat::CompletionCreateParams.fields.keys.uniq.freeze
 
-        def initialize(client:)
-          @client = client
-        end
-
         def create(message: "", parameters: {})
           validate_and_resolve!(parameters, message)
           perform_request(parameters:)
         end
 
         def perform_request(parameters: {})
+          add_request_headers(parameters, REQUESTED_MODEL_HEADER => parameters[:model])
           @client.chat.completions.create(**parameters.slice(*ALLOWED_EXTRA_PARAMETERS))
         end
 
@@ -29,7 +27,6 @@ module NitroIntelligence
             metadata: {},
             messages: [],
             model: NitroIntelligence.model_catalog.default_text_model&.name,
-            extra_headers: { "Prefer" => "wait" },
           }
 
           parameters.replace(default_parameters.merge(parameters))

data/lib/nitro_intelligence/client/handlers/image_handler.rb

@@ -1,16 +1,13 @@
 require "openai"
+require "nitro_intelligence/client/handlers/base_handler"
 require "nitro_intelligence/media/image_generation"
 
 module NitroIntelligence
   module Client
     module Handlers
-      class ImageHandler
+      class ImageHandler < BaseHandler
         ALLOWED_EXTRA_PARAMETERS = OpenAI::Models::Chat::CompletionCreateParams.fields.keys.uniq.freeze
 
-        def initialize(client:)
-          @client = client
-        end
-
         def create(message: "", target_image: nil, reference_images: [], parameters: {})
           image_generation = build_image_generation(message:, target_image:, reference_images:, parameters:)
 
@@ -23,6 +20,7 @@ module NitroIntelligence
         end
 
         def perform_request(parameters: {})
+          add_request_headers(parameters, MODALITY_HEADER => "image", REQUESTED_MODEL_HEADER => parameters[:model])
           @client.chat.completions.create(**parameters.slice(*ALLOWED_EXTRA_PARAMETERS))
         end
 
@@ -32,7 +30,6 @@ module NitroIntelligence
             metadata: {},
             messages: image_generation.messages,
             model: image_generation.config.model,
-            extra_headers: { "Prefer" => "wait" },
             request_options: {
               extra_body: {
                 image_config: {

data/lib/nitro_intelligence/client/handlers/text_to_speech_handler.rb

@@ -1,21 +1,19 @@
 require "openai"
+require "nitro_intelligence/client/handlers/base_handler"
 
 module NitroIntelligence
   module Client
     module Handlers
-      class TextToSpeechHandler
+      class TextToSpeechHandler < BaseHandler
         ALLOWED_EXTRA_PARAMETERS = OpenAI::Models::Audio::SpeechCreateParams.fields.keys.uniq.freeze
 
-        def initialize(client:)
-          @client = client
-        end
-
         def create(message: "", parameters: {})
           validate_and_resolve!(parameters)
           perform_request(message:, parameters:)
         end
 
         def perform_request(message: "", parameters: {})
+          add_request_headers(parameters, MODALITY_HEADER => "audio", REQUESTED_MODEL_HEADER => parameters[:model])
           @client.audio.speech.create(
             input: message,
             **parameters.slice(*ALLOWED_EXTRA_PARAMETERS)
@@ -25,8 +23,6 @@ module NitroIntelligence
         def validate_and_resolve!(parameters)
           model_name = parameters[:model] || NitroIntelligence.model_catalog.default_text_to_speech_model&.name
           model = NitroIntelligence.model_catalog.lookup_by_name(model_name)
-
-          # Check model supported
           raise ArgumentError, "Unsupported model: '#{model_name}'" unless model
 
           default_parameters = {
@@ -35,17 +31,19 @@ module NitroIntelligence
             voice: model.default_voice,
             response_format: model.default_response_format,
           }
-
           parameters.replace(default_parameters.merge(parameters))
+          validate_voice_and_format!(model, parameters)
+        end
+
+      private
 
-          # Check voice supported
+        def validate_voice_and_format!(model, parameters)
           unless model.voices.include?(parameters[:voice])
             raise ArgumentError,
                   "Unsupported voice: '#{parameters[:voice]}'. " \
                   "Supported voices for #{model.name} are: #{model.voices}"
           end
 
-          # Check format supported
           return if model.response_formats.include?(parameters[:response_format])
 
           raise ArgumentError,

data/lib/nitro_intelligence/version.rb

@@ -1,3 +1,3 @@
 module NitroIntelligence
-  VERSION = "2.0.0".freeze
+  VERSION = "2.1.0".freeze
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: nitro_intelligence
 version: !ruby/object:Gem::Version
-  version: 2.0.0
+  version: 2.1.0
 platform: ruby
 authors:
 - Igor Artemenko
@@ -80,6 +80,7 @@ files:
 - lib/nitro_intelligence/client/client.rb
 - lib/nitro_intelligence/client/factory.rb
 - lib/nitro_intelligence/client/handlers/audio_transcription_handler.rb
+- lib/nitro_intelligence/client/handlers/base_handler.rb
 - lib/nitro_intelligence/client/handlers/chat_handler.rb
 - lib/nitro_intelligence/client/handlers/image_handler.rb
 - lib/nitro_intelligence/client/handlers/observed/audio_transcription_handler.rb