remove wrong migration file

Gemfile

@@ -31,6 +31,14 @@ gem 'haikunator'
 gem 'liquid'
 # Parse Markdown to HTML
 gem 'commonmarker'
+# Convert PDF pages to images for content extraction
+gem 'pdftoimage'
+# Google Vision API for text extraction from images
+gem 'google-cloud-vision'
+# Tesseract OCR fallback when Vision API not available
+gem 'rtesseract'
+# Image manipulation for PDF page batching
+gem 'mini_magick'
 # Validate Data against JSON Schema
 gem 'json_schemer'
 # used in swagger build

Gemfile.lock

@@ -275,14 +275,16 @@ GEM
     fugit (1.11.1)
       et-orbi (~> 1, >= 1.2.11)
       raabro (~> 1.4)
-    gapic-common (0.20.0)
+    gapic-common (1.0.0)
       faraday (>= 1.9, < 3.a)
       faraday-retry (>= 1.0, < 3.a)
-      google-protobuf (~> 3.14)
+      google-cloud-env (~> 2.2)
-      googleapis-common-protos (>= 1.3.12, < 2.a)
+      google-logging-utils (~> 0.1)
-      googleapis-common-protos-types (>= 1.3.1, < 2.a)
+      google-protobuf (>= 3.25, < 5.a)
-      googleauth (~> 1.0)
+      googleapis-common-protos (~> 1.6)
-      grpc (~> 1.36)
+      googleapis-common-protos-types (~> 1.15)
+      googleauth (~> 1.12)
+      grpc (~> 1.66)
     geocoder (1.8.1)
     gli (2.22.2)
       ostruct
@@ -327,6 +329,13 @@ GEM
     google-cloud-translate-v3 (0.10.0)
       gapic-common (>= 0.20.0, < 2.a)
       google-cloud-errors (~> 1.0)
+    google-cloud-vision (2.0.2)
+      google-cloud-core (~> 1.6)
+      google-cloud-vision-v1 (~> 1.2)
+    google-cloud-vision-v1 (1.3.0)
+      gapic-common (~> 1.0)
+      google-cloud-errors (~> 1.0)
+    google-logging-utils (0.2.0)
     google-protobuf (3.25.7)
     googleapis-common-protos (1.6.0)
       google-protobuf (>= 3.18, < 5.a)
@@ -334,9 +343,10 @@ GEM
       grpc (~> 1.41)
     googleapis-common-protos-types (1.20.0)
       google-protobuf (>= 3.18, < 5.a)
-    googleauth (1.11.2)
+    googleauth (1.14.0)
       faraday (>= 1.0, < 3.a)
-      google-cloud-env (~> 2.1)
+      google-cloud-env (~> 2.2)
+      google-logging-utils (~> 0.1)
       jwt (>= 1.4, < 3.0)
       multi_json (~> 1.11)
       os (>= 0.9, < 2.0)
@@ -381,6 +391,7 @@ GEM
     httpclient (2.8.3)
     i18n (1.14.7)
       concurrent-ruby (~> 1.0)
+    iconv (1.1.0)
     image_processing (1.12.2)
       mini_magick (>= 4.9.5, < 5)
       ruby-vips (>= 2.0.17, < 3)
@@ -550,6 +561,9 @@ GEM
     parser (3.3.8.0)
       ast (~> 2.4.1)
       racc
+    pdftoimage (0.2.1)
+      iconv (~> 1.0)
+      shellwords (~> 0.2.2)
     pg (1.5.3)
     pg_search (2.3.6)
       activerecord (>= 5.2)
@@ -672,6 +686,7 @@ GEM
     rspec-support (3.13.1)
     rspec_junit_formatter (0.6.0)
       rspec-core (>= 2, < 4, != 2.12.0)
+    rtesseract (3.1.4)
     rubocop (1.75.6)
       json (~> 2.3)
       language_server-protocol (~> 3.17.0.2)
@@ -747,6 +762,7 @@ GEM
       sentry-ruby (~> 5.19.0)
       sidekiq (>= 3.0)
     sexp_processor (4.17.0)
+    shellwords (0.2.2)
     shopify_api (14.8.0)
       activesupport
       concurrent-ruby
@@ -927,6 +943,7 @@ DEPENDENCIES
   google-cloud-dialogflow-v2 (>= 0.24.0)
   google-cloud-storage (>= 1.48.0)
   google-cloud-translate-v3 (>= 0.7.0)
+  google-cloud-vision
   groupdate
   grpc
   haikunator
@@ -950,6 +967,7 @@ DEPENDENCIES
   lograge (~> 0.14.0)
   maxminddb
   meta_request (>= 0.8.3)
+  mini_magick
   mock_redis
   neighbor
   net-smtp (~> 0.3.4)
@@ -959,6 +977,7 @@ DEPENDENCIES
   omniauth-google-oauth2 (>= 1.1.3)
   omniauth-oauth2
   omniauth-rails_csrf_protection (~> 1.0, >= 1.0.2)
+  pdftoimage
   pg
   pg_search
   pgvector
@@ -978,6 +997,7 @@ DEPENDENCIES
   reverse_markdown
   rspec-rails (>= 6.1.5)
   rspec_junit_formatter
+  rtesseract
   rubocop
   rubocop-factory_bot
   rubocop-performance
@@ -1020,4 +1040,4 @@ RUBY VERSION
    ruby 3.4.4p34
 
 BUNDLED WITH
-   2.5.16
+   2.6.7

db/migrate/20250623120000_add_document_type_and_file_to_captain_documents.rb

@@ -0,0 +1,6 @@
+class AddDocumentTypeAndFileToCaptainDocuments < ActiveRecord::Migration[7.0]
+  def change
+    add_column :captain_documents, :document_type, :integer, default: 0, null: false
+    add_index :captain_documents, :document_type
+  end
+end

db/schema.rb

@@ -10,7 +10,7 @@
 #
 # It's strongly recommended that you check this file into your version control system.
 
-ActiveRecord::Schema[7.1].define(version: 2025_05_23_031839) do
+ActiveRecord::Schema[7.1].define(version: 2025_06_23_120000) do
   # These extensions should be enabled to support this database
   enable_extension "pg_stat_statements"
   enable_extension "pg_trgm"
@@ -289,9 +289,11 @@ ActiveRecord::Schema[7.1].define(version: 2025_05_23_031839) do
     t.datetime "created_at", null: false
     t.datetime "updated_at", null: false
     t.integer "status", default: 0, null: false
+    t.integer "document_type", default: 0, null: false
     t.index ["account_id"], name: "index_captain_documents_on_account_id"
     t.index ["assistant_id", "external_link"], name: "index_captain_documents_on_assistant_id_and_external_link", unique: true
     t.index ["assistant_id"], name: "index_captain_documents_on_assistant_id"
+    t.index ["document_type"], name: "index_captain_documents_on_document_type"
     t.index ["status"], name: "index_captain_documents_on_status"
   end
 
@@ -907,7 +909,7 @@ ActiveRecord::Schema[7.1].define(version: 2025_05_23_031839) do
     t.text "header_text"
     t.datetime "created_at", null: false
     t.datetime "updated_at", null: false
-    t.jsonb "config", default: {"allowed_locales"=>["en"]}
+    t.jsonb "config", default: {"allowed_locales" => ["en"]}
     t.boolean "archived", default: false
     t.bigint "channel_web_widget_id"
     t.index ["channel_web_widget_id"], name: "index_portals_on_channel_web_widget_id"

docker/Dockerfile

@@ -34,6 +34,8 @@ RUN apk update && apk add --no-cache \
   git \
   curl \
   xz \
+  poppler-utils \
+  tesseract-ocr \
   && mkdir -p /var/app \
   && gem install bundler
 
@@ -128,6 +130,8 @@ RUN apk update && apk add --no-cache \
   imagemagick \
   git \
   vips \
+  poppler-utils \
+  tesseract-ocr \
   && gem install bundler
 
 COPY --from=node /usr/local/bin/node /usr/local/bin/

enterprise/app/controllers/api/v1/accounts/captain/assistants_controller.rb

@@ -25,8 +25,8 @@ class Api::V1::Accounts::Captain::AssistantsController < Api::V1::Accounts::Base
 
   def playground
     response = Captain::Llm::AssistantChatService.new(assistant: @assistant).generate_response(
-      params[:message_content],
+      additional_message: params[:message_content],
-      message_history
+      message_history: message_history
     )
 
     render json: response

enterprise/app/jobs/captain/conversation/response_builder_job.rb

@@ -26,8 +26,7 @@ class Captain::Conversation::ResponseBuilderJob < ApplicationJob
 
   def generate_and_process_response
     @response = Captain::Llm::AssistantChatService.new(assistant: @assistant).generate_response(
-      @conversation.messages.incoming.last.content,
+      message_history: collect_previous_messages
-      collect_previous_messages
     )
 
     return process_action('handoff') if handoff_requested?
@@ -43,33 +42,11 @@ class Captain::Conversation::ResponseBuilderJob < ApplicationJob
       .where(message_type: [:incoming, :outgoing])
       .where(private: false)
       .map do |message|
-        {
+      {
-          content: message_content(message),
+        content: prepare_multimodal_message_content(message),
-          role: determine_role(message)
+        role: determine_role(message)
-        }
+      }
-      end
-  end
-
-  def message_content(message)
-    return message.content if message.content.present?
-    return 'User has shared a message without content' unless message.attachments.any?
-
-    audio_transcriptions = extract_audio_transcriptions(message.attachments)
-    return audio_transcriptions if audio_transcriptions.present?
-
-    'User has shared an attachment'
-  end
-
-  def extract_audio_transcriptions(attachments)
-    audio_attachments = attachments.where(file_type: :audio)
-    return '' if audio_attachments.blank?
-
-    transcriptions = ''
-    audio_attachments.each do |attachment|
-      result = Messages::AudioTranscriptionService.new(attachment).perform
-      transcriptions += result[:transcriptions] if result[:success]
     end
-    transcriptions
   end
 
   def determine_role(message)
@@ -78,6 +55,10 @@ class Captain::Conversation::ResponseBuilderJob < ApplicationJob
     message.message_type == 'incoming' ? 'user' : 'system'
   end
 
+  def prepare_multimodal_message_content(message)
+    Captain::OpenAiMessageBuilderService.new(message: message).generate_content
+  end
+
   def handoff_requested?
     @response['response'] == 'conversation_handoff'
   end

enterprise/app/jobs/captain/documents/pdf_processing_job.rb

@@ -0,0 +1,30 @@
+class Captain::Documents::PdfProcessingJob < ApplicationJob
+  queue_as :low
+
+  def perform(document)
+    return unless document.pdf_document?
+
+    Rails.logger.info "Starting PDF processing for document #{document.id}"
+    
+    begin
+      service = Captain::Tools::PdfProcessingService.new(document)
+      extracted_content = service.process
+      
+      Rails.logger.info "Successfully processed PDF document #{document.id}, extracted #{extracted_content.length} characters"
+      
+    rescue Captain::Tools::PdfProcessingService::PdfProcessingError => e
+      Rails.logger.error "PDF processing failed for document #{document.id}: #{e.message}"
+      
+      # Mark document as failed (you may want to add a 'failed' status to the enum)
+      document.update!(status: :in_progress) # Keep as in_progress for potential retry
+      
+      # Optionally, you could add a retry mechanism or notification
+      raise e
+      
+    rescue StandardError => e
+      Rails.logger.error "Unexpected error processing PDF document #{document.id}: #{e.message}"
+      document.update!(status: :in_progress)
+      raise e
+    end
+  end
+end

enterprise/app/models/captain/document.rb

@@ -4,6 +4,7 @@
 #
 #  id            :bigint           not null, primary key
 #  content       :text
+#  document_type :integer          default("url"), not null
 #  external_link :string           not null
 #  name          :string
 #  status        :integer          default("in_progress"), not null
@@ -17,6 +18,7 @@
 #  index_captain_documents_on_account_id                      (account_id)
 #  index_captain_documents_on_assistant_id                    (assistant_id)
 #  index_captain_documents_on_assistant_id_and_external_link  (assistant_id,external_link) UNIQUE
+#  index_captain_documents_on_document_type                   (document_type)
 #  index_captain_documents_on_status                          (status)
 #
 class Captain::Document < ApplicationRecord
@@ -37,7 +39,13 @@ class Captain::Document < ApplicationRecord
     available: 1
   }
 
+  enum document_type: {
+    url: 0,
+    pdf: 1
+  }
+
   before_create :ensure_within_plan_limit
+  before_create :detect_document_type
   after_create_commit :enqueue_crawl_job
   after_create_commit :update_document_usage
   after_destroy :update_document_usage
@@ -47,12 +55,38 @@ class Captain::Document < ApplicationRecord
   scope :for_account, ->(account_id) { where(account_id: account_id) }
   scope :for_assistant, ->(assistant_id) { where(assistant_id: assistant_id) }
 
+  def url_document?
+    document_type == 'url'
+  end
+
+  def pdf_document?
+    document_type == 'pdf'
+  end
+
   private
 
   def enqueue_crawl_job
     return if status != 'in_progress'
 
-    Captain::Documents::CrawlJob.perform_later(self)
+    if pdf_document?
+      Captain::Documents::PdfProcessingJob.perform_later(self)
+    else
+      Captain::Documents::CrawlJob.perform_later(self)
+    end
+  end
+
+  def detect_document_type
+    return unless external_link.present?
+    
+    # Check if URL points to a PDF file
+    uri = URI.parse(external_link)
+    if uri.path.downcase.end_with?('.pdf') || external_link.downcase.include?('.pdf')
+      self.document_type = :pdf
+    else
+      self.document_type = :url
+    end
+  rescue URI::InvalidURIError
+    self.document_type = :url
   end
 
   def enqueue_response_builder_job

enterprise/app/services/captain/llm/assistant_chat_service.rb

@@ -12,9 +12,16 @@ class Captain::Llm::AssistantChatService < Llm::BaseOpenAiService
     register_tools
   end
 
-  def generate_response(input, previous_messages = [], role = 'user')
+  # additional_message: A single message (String) from the user that should be appended to the chat.
-    @messages += previous_messages
+  #                    It can be an empty String or nil when you only want to supply historical messages.
-    @messages << { role: role, content: input } if input.present?
+  # message_history:   An Array of already formatted messages that provide the previous context.
+  # role:              The role for the additional_message (defaults to `user`).
+  #
+  # NOTE: Parameters are provided as keyword arguments to improve clarity and avoid relying on
+  # positional ordering.
+  def generate_response(additional_message: nil, message_history: [], role: 'user')
+    @messages += message_history
+    @messages << { role: role, content: additional_message } if additional_message.present?
     request_chat_completion
   end
 

enterprise/app/services/captain/open_ai_message_builder_service.rb

@@ -0,0 +1,59 @@
+class Captain::OpenAiMessageBuilderService
+  pattr_initialize [:message!]
+
+  def generate_content
+    parts = []
+    parts << text_part(@message.content) if @message.content.present?
+    parts.concat(attachment_parts(@message.attachments)) if @message.attachments.any?
+
+    return 'Message without content' if parts.blank?
+    return parts.first[:text] if parts.one? && parts.first[:type] == 'text'
+
+    parts
+  end
+
+  private
+
+  def text_part(text)
+    { type: 'text', text: text }
+  end
+
+  def image_part(image_url)
+    { type: 'image_url', image_url: { url: image_url } }
+  end
+
+  def attachment_parts(attachments)
+    image_attachments = attachments.where(file_type: :image)
+    image_content = image_parts(image_attachments)
+
+    transcription = extract_audio_transcriptions(attachments)
+    transcription_part = text_part(transcription) if transcription.present?
+
+    attachment_part = text_part('User has shared an attachment') if attachments.where.not(file_type: %i[image audio]).exists?
+
+    [image_content, transcription_part, attachment_part].flatten.compact
+  end
+
+  def image_parts(image_attachments)
+    image_attachments.each_with_object([]) do |attachment, parts|
+      url = get_attachment_url(attachment)
+      parts << image_part(url) if url.present?
+    end
+  end
+
+  def get_attachment_url(attachment)
+    return attachment.external_url if attachment.external_url.present?
+
+    attachment.file.attached? ? attachment.file_url : nil
+  end
+
+  def extract_audio_transcriptions(attachments)
+    audio_attachments = attachments.where(file_type: :audio)
+    return '' if audio_attachments.blank?
+
+    audio_attachments.map do |attachment|
+      result = Messages::AudioTranscriptionService.new(attachment).perform
+      result[:success] ? result[:transcriptions] : ''
+    end.join
+  end
+end

enterprise/app/services/captain/tools/pdf_page_batcher_service.rb

@@ -0,0 +1,57 @@
+require 'mini_magick'
+
+class Captain::Tools::PdfPageBatcherService
+  PAGES_PER_BATCH = 3
+  
+  def initialize(temp_files_tracker)
+    @temp_files = temp_files_tracker
+  end
+
+  def batch_pages(image_files)
+    return image_files if image_files.length <= PAGES_PER_BATCH
+    
+    batched_images = []
+    
+    image_files.each_slice(PAGES_PER_BATCH) do |batch|
+      if batch.length == 1
+        # Single page, use as-is
+        batched_images << batch.first
+      else
+        # Multiple pages, combine them
+        combined_image = combine_images(batch)
+        batched_images << combined_image
+      end
+    end
+    
+    batched_images
+  end
+
+  private
+
+  def combine_images(image_paths)
+    # Create a vertical montage of the images
+    output_path = create_temp_file_path('.png')
+    
+    MiniMagick::Tool::Montage.new do |montage|
+      image_paths.each { |path| montage << path }
+      montage.tile('1x')  # Arrange in single column (vertically)
+      montage.geometry('+0+10')  # Add 10px spacing between images
+      montage.background('white')
+      montage << output_path
+    end
+    
+    @temp_files << output_path
+    output_path
+  rescue StandardError => e
+    Rails.logger.error "Failed to combine images: #{e.message}"
+    # Fallback to first image if combining fails
+    image_paths.first
+  end
+
+  def create_temp_file_path(extension)
+    temp_file = Tempfile.new(['combined_pages', extension])
+    temp_file.close
+    @temp_files << temp_file
+    temp_file.path
+  end
+end

enterprise/app/services/captain/tools/pdf_processing_service.rb

@@ -0,0 +1,75 @@
+class Captain::Tools::PdfProcessingService
+  def initialize(document)
+    @document = document
+    @temp_files = []
+  end
+
+  def process
+    return unless @document.pdf_document?
+
+    content = extract_text_content
+    
+    # Follow the same pattern as SimplePageCrawlParserJob
+    @document.update!(
+      content: content[0..14_999], # Same limit as crawler
+      status: :available
+    )
+    
+    content
+  rescue StandardError => e
+    Rails.logger.error "PDF processing failed for document #{@document.id}: #{e.message}"
+    @document.update!(content: "Error processing PDF: #{e.message}", status: :available)
+    raise e
+  ensure
+    cleanup_temp_files
+  end
+
+  private
+
+  def extract_text_content
+    require 'down'
+    
+    pdf_file = Down.download(@document.external_link)
+    @temp_files << pdf_file
+    
+    output_dir = Dir.mktmpdir
+    @temp_files << output_dir
+    
+    # Convert PDF to images
+    system("pdftoimage", "-png", "-r", "150", pdf_file.path, "#{output_dir}/page")
+    
+    image_files = Dir.glob("#{output_dir}/page-*.png").sort
+    return fallback_content if image_files.empty?
+    
+    # Batch pages for efficient processing
+    batcher = Captain::Tools::PdfPageBatcherService.new(@temp_files)
+    batched_images = batcher.batch_pages(image_files)
+    
+    # Extract text using Google Vision
+    vision_extractor = Captain::Tools::VisionTextExtractorService.new
+    extracted_text = vision_extractor.extract_text_from_multiple_images(batched_images)
+    
+    extracted_text.present? ? extracted_text : fallback_content
+  end
+
+  def fallback_content
+    "PDF Document: #{@document.name || 'Untitled'}\nSource: #{@document.external_link}"
+  end
+
+  def cleanup_temp_files
+    @temp_files.each do |file_or_dir|
+      case file_or_dir
+      when String # directory path
+        FileUtils.rm_rf(file_or_dir)
+      when Tempfile
+        file_or_dir.close
+        file_or_dir.unlink
+      else # Down::ChunkedIO or similar
+        file_or_dir.close if file_or_dir.respond_to?(:close)
+        file_or_dir.unlink if file_or_dir.respond_to?(:unlink)
+      end
+    rescue StandardError => e
+      Rails.logger.warn "Failed to cleanup temp file: #{e.message}"
+    end
+  end
+end

enterprise/app/services/captain/tools/vision_text_extractor_service.rb

@@ -0,0 +1,68 @@
+class Captain::Tools::VisionTextExtractorService
+  class TextExtractionError < StandardError; end
+
+  def initialize
+    @use_vision = vision_available?
+    @vision = Google::Cloud::Vision.image_annotator if @use_vision
+  end
+
+  def extract_text_from_image(image_path)
+    if @use_vision
+      extract_with_vision(image_path)
+    else
+      extract_with_tesseract(image_path)
+    end
+  end
+
+  def extract_text_from_multiple_images(image_paths)
+    Rails.logger.info "Using #{@use_vision ? 'Google Vision' : 'Tesseract'} for text extraction"
+    
+    extracted_texts = []
+    
+    image_paths.each_with_index do |image_path, index|
+      Rails.logger.info "Extracting text from image #{index + 1}/#{image_paths.length}"
+      
+      text = extract_text_from_image(image_path)
+      extracted_texts << text if text.present?
+    end
+    
+    extracted_texts.join("\n\n")
+  end
+
+  private
+
+  def vision_available?
+    require 'google/cloud/vision'
+    Google::Cloud::Vision.image_annotator
+    true
+  rescue StandardError => e
+    Rails.logger.info "Google Vision not available, falling back to Tesseract: #{e.message}"
+    false
+  end
+
+  def extract_with_vision(image_path)
+    image = Google::Cloud::Vision::Image.new(image_path)
+    response = @vision.text_detection(image: image)
+    
+    if response.error
+      raise TextExtractionError, "Vision API error: #{response.error.message}"
+    end
+    
+    text_annotation = response.full_text_annotation
+    return "" unless text_annotation
+    
+    text_annotation.text
+  rescue Google::Cloud::Error => e
+    Rails.logger.error "Google Vision API error: #{e.message}"
+    raise TextExtractionError, "Failed to extract text: #{e.message}"
+  end
+
+  def extract_with_tesseract(image_path)
+    require 'rtesseract'
+    
+    RTesseract.new(image_path).to_s.strip
+  rescue StandardError => e
+    Rails.logger.error "Tesseract extraction error: #{e.message}"
+    raise TextExtractionError, "Failed to extract text with Tesseract: #{e.message}"
+  end
+end

enterprise/app/views/api/v1/models/captain/_document.json.jbuilder

@@ -4,6 +4,7 @@ json.assistant do
 end
 json.content resource.content
 json.created_at resource.created_at.to_i
+json.document_type resource.document_type
 json.external_link resource.external_link
 json.id resource.id
 json.name resource.name

spec/enterprise/controllers/api/v1/accounts/captain/assistants_controller_spec.rb

@@ -211,8 +211,8 @@ RSpec.describe 'Api::V1::Accounts::Captain::Assistants', type: :request do
 
         expect(response).to have_http_status(:success)
         expect(chat_service).to have_received(:generate_response).with(
-          valid_params[:message_content],
+          additional_message: valid_params[:message_content],
-          valid_params[:message_history]
+          message_history: valid_params[:message_history]
         )
         expect(json_response[:content]).to eq('Assistant response')
       end
@@ -232,8 +232,8 @@ RSpec.describe 'Api::V1::Accounts::Captain::Assistants', type: :request do
 
         expect(response).to have_http_status(:success)
         expect(chat_service).to have_received(:generate_response).with(
-          params_without_history[:message_content],
+          additional_message: params_without_history[:message_content],
-          []
+          message_history: []
         )
       end
     end

spec/enterprise/jobs/captain/conversation/response_builder_job_spec.rb

@@ -30,5 +30,30 @@ RSpec.describe Captain::Conversation::ResponseBuilderJob, type: :job do
       account.reload
       expect(account.usage_limits[:captain][:responses][:consumed]).to eq(1)
     end
+
+    context 'when message contains an image' do
+      let(:message_with_image) { create(:message, conversation: conversation, message_type: :incoming, content: 'Can you help with this error?') }
+      let(:image_attachment) { message_with_image.attachments.create!(account: account, file_type: :image, external_url: 'https://example.com/error.jpg') }
+
+      before do
+        image_attachment
+      end
+
+      it 'includes image URL directly in the message content for OpenAI vision analysis' do
+        # Expect the generate_response to receive multimodal content with image URL
+        expect(mock_llm_chat_service).to receive(:generate_response) do |**kwargs|
+          history = kwargs[:message_history]
+          last_entry = history.last
+          expect(last_entry[:content]).to be_an(Array)
+          expect(last_entry[:content].any? { |part| part[:type] == 'text' && part[:text] == 'Can you help with this error?' }).to be true
+          expect(last_entry[:content].any? do |part|
+            part[:type] == 'image_url' && part[:image_url][:url] == 'https://example.com/error.jpg'
+          end).to be true
+          { 'response' => 'I can see the error in your image. It appears to be a database connection issue.' }
+        end
+
+        described_class.perform_now(conversation, assistant)
+      end
+    end
   end
 end

spec/enterprise/models/captain/document_spec.rb

@@ -0,0 +1,69 @@
+require 'rails_helper'
+
+RSpec.describe Captain::Document, type: :model do
+  let(:account) { create(:account) }
+  let(:assistant) { create(:captain_assistant, account: account) }
+
+  describe 'document_type detection' do
+    it 'detects PDF documents from URL' do
+      document = assistant.documents.build(
+        name: 'Test PDF',
+        external_link: 'https://example.com/document.pdf'
+      )
+      
+      document.save!
+      
+      expect(document.document_type).to eq('pdf')
+      expect(document.pdf_document?).to be true
+      expect(document.url_document?).to be false
+    end
+
+    it 'detects URL documents for non-PDF links' do
+      document = assistant.documents.build(
+        name: 'Test Page',
+        external_link: 'https://example.com/page.html'
+      )
+      
+      document.save!
+      
+      expect(document.document_type).to eq('url')
+      expect(document.url_document?).to be true
+      expect(document.pdf_document?).to be false
+    end
+
+    it 'defaults to URL for invalid URIs' do
+      document = assistant.documents.build(
+        name: 'Invalid URL',
+        external_link: 'not-a-valid-url'
+      )
+      
+      document.save!
+      
+      expect(document.document_type).to eq('url')
+    end
+  end
+
+  describe 'job enqueueing' do
+    it 'enqueues PDF processing job for PDF documents' do
+      document = assistant.documents.build(
+        name: 'Test PDF',
+        external_link: 'https://example.com/document.pdf'
+      )
+      
+      expect(Captain::Documents::PdfProcessingJob).to receive(:perform_later).with(document)
+      
+      document.save!
+    end
+
+    it 'enqueues crawl job for URL documents' do
+      document = assistant.documents.build(
+        name: 'Test Page',
+        external_link: 'https://example.com/page.html'
+      )
+      
+      expect(Captain::Documents::CrawlJob).to receive(:perform_later).with(document)
+      
+      document.save!
+    end
+  end
+end

spec/enterprise/services/captain/open_ai_message_builder_service_spec.rb

@@ -0,0 +1,309 @@
+require 'rails_helper'
+
+RSpec.describe Captain::OpenAiMessageBuilderService do
+  subject(:service) { described_class.new(message: message) }
+
+  let(:message) { create(:message, content: 'Hello world') }
+
+  describe '#generate_content' do
+    context 'when message has only text content' do
+      it 'returns the text content directly' do
+        expect(service.generate_content).to eq('Hello world')
+      end
+    end
+
+    context 'when message has no content and no attachments' do
+      let(:message) { create(:message, content: nil) }
+
+      it 'returns default message' do
+        expect(service.generate_content).to eq('Message without content')
+      end
+    end
+
+    context 'when message has text content and attachments' do
+      before do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :image, external_url: 'https://example.com/image.jpg')
+        attachment.save!
+      end
+
+      it 'returns an array of content parts' do
+        result = service.generate_content
+        expect(result).to be_an(Array)
+        expect(result).to include({ type: 'text', text: 'Hello world' })
+        expect(result).to include({ type: 'image_url', image_url: { url: 'https://example.com/image.jpg' } })
+      end
+    end
+
+    context 'when message has only non-text attachments' do
+      let(:message) { create(:message, content: nil) }
+
+      before do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :image, external_url: 'https://example.com/image.jpg')
+        attachment.save!
+      end
+
+      it 'returns an array of content parts without text' do
+        result = service.generate_content
+        expect(result).to be_an(Array)
+        expect(result).to include({ type: 'image_url', image_url: { url: 'https://example.com/image.jpg' } })
+        expect(result).not_to include(hash_including(type: 'text', text: 'Hello world'))
+      end
+    end
+  end
+
+  describe '#attachment_parts' do
+    let(:message) { create(:message, content: nil) }
+    let(:attachments) { message.attachments }
+
+    context 'with image attachments' do
+      before do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :image, external_url: 'https://example.com/image.jpg')
+        attachment.save!
+      end
+
+      it 'includes image parts' do
+        result = service.send(:attachment_parts, attachments)
+        expect(result).to include({ type: 'image_url', image_url: { url: 'https://example.com/image.jpg' } })
+      end
+    end
+
+    context 'with audio attachments' do
+      let(:audio_attachment) do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :audio)
+        attachment.save!
+        attachment
+      end
+
+      before do
+        allow(Messages::AudioTranscriptionService).to receive(:new).with(audio_attachment).and_return(
+          instance_double(Messages::AudioTranscriptionService, perform: { success: true, transcriptions: 'Audio transcription text' })
+        )
+      end
+
+      it 'includes transcription text part' do
+        audio_attachment # trigger creation
+        result = service.send(:attachment_parts, attachments)
+        expect(result).to include({ type: 'text', text: 'Audio transcription text' })
+      end
+    end
+
+    context 'with other file types' do
+      before do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :file)
+        attachment.save!
+      end
+
+      it 'includes generic attachment message' do
+        result = service.send(:attachment_parts, attachments)
+        expect(result).to include({ type: 'text', text: 'User has shared an attachment' })
+      end
+    end
+
+    context 'with mixed attachment types' do
+      let(:image_attachment) do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :image, external_url: 'https://example.com/image.jpg')
+        attachment.save!
+        attachment
+      end
+
+      let(:audio_attachment) do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :audio)
+        attachment.save!
+        attachment
+      end
+
+      let(:document_attachment) do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :file)
+        attachment.save!
+        attachment
+      end
+
+      before do
+        allow(Messages::AudioTranscriptionService).to receive(:new).with(audio_attachment).and_return(
+          instance_double(Messages::AudioTranscriptionService, perform: { success: true, transcriptions: 'Audio text' })
+        )
+      end
+
+      it 'includes all relevant parts' do
+        image_attachment    # trigger creation
+        audio_attachment    # trigger creation
+        document_attachment # trigger creation
+
+        result = service.send(:attachment_parts, attachments)
+        expect(result).to include({ type: 'image_url', image_url: { url: 'https://example.com/image.jpg' } })
+        expect(result).to include({ type: 'text', text: 'Audio text' })
+        expect(result).to include({ type: 'text', text: 'User has shared an attachment' })
+      end
+    end
+  end
+
+  describe '#image_parts' do
+    let(:message) { create(:message, content: nil) }
+
+    context 'with valid image attachments' do
+      let(:image1) do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :image, external_url: 'https://example.com/image1.jpg')
+        attachment.save!
+        attachment
+      end
+
+      let(:image2) do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :image, external_url: 'https://example.com/image2.jpg')
+        attachment.save!
+        attachment
+      end
+
+      it 'returns image parts for all valid images' do
+        image1 # trigger creation
+        image2 # trigger creation
+
+        image_attachments = message.attachments.where(file_type: :image)
+        result = service.send(:image_parts, image_attachments)
+
+        expect(result).to include({ type: 'image_url', image_url: { url: 'https://example.com/image1.jpg' } })
+        expect(result).to include({ type: 'image_url', image_url: { url: 'https://example.com/image2.jpg' } })
+      end
+    end
+
+    context 'with image attachments without URLs' do
+      let(:image_attachment) do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :image, external_url: nil)
+        attachment.save!
+        attachment
+      end
+
+      before do
+        allow(image_attachment).to receive(:file).and_return(instance_double(ActiveStorage::Attached::One, attached?: false))
+      end
+
+      it 'skips images without valid URLs' do
+        image_attachment # trigger creation
+
+        image_attachments = message.attachments.where(file_type: :image)
+        result = service.send(:image_parts, image_attachments)
+
+        expect(result).to be_empty
+      end
+    end
+  end
+
+  describe '#get_attachment_url' do
+    let(:attachment) do
+      attachment = message.attachments.build(account_id: message.account_id, file_type: :image)
+      attachment.save!
+      attachment
+    end
+
+    context 'when attachment has external_url' do
+      before { attachment.update(external_url: 'https://example.com/image.jpg') }
+
+      it 'returns external_url' do
+        expect(service.send(:get_attachment_url, attachment)).to eq('https://example.com/image.jpg')
+      end
+    end
+
+    context 'when attachment has attached file' do
+      before do
+        attachment.update(external_url: nil)
+        allow(attachment).to receive(:file).and_return(instance_double(ActiveStorage::Attached::One, attached?: true))
+        allow(attachment).to receive(:file_url).and_return('https://local.com/file.jpg')
+      end
+
+      it 'returns file_url' do
+        expect(service.send(:get_attachment_url, attachment)).to eq('https://local.com/file.jpg')
+      end
+    end
+
+    context 'when attachment has no URL or file' do
+      before do
+        attachment.update(external_url: nil)
+        allow(attachment).to receive(:file).and_return(instance_double(ActiveStorage::Attached::One, attached?: false))
+      end
+
+      it 'returns nil' do
+        expect(service.send(:get_attachment_url, attachment)).to be_nil
+      end
+    end
+  end
+
+  describe '#extract_audio_transcriptions' do
+    let(:message) { create(:message, content: nil) }
+
+    context 'with no audio attachments' do
+      it 'returns empty string' do
+        result = service.send(:extract_audio_transcriptions, message.attachments)
+        expect(result).to eq('')
+      end
+    end
+
+    context 'with successful audio transcriptions' do
+      let(:audio1) do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :audio)
+        attachment.save!
+        attachment
+      end
+
+      let(:audio2) do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :audio)
+        attachment.save!
+        attachment
+      end
+
+      before do
+        allow(Messages::AudioTranscriptionService).to receive(:new).with(audio1).and_return(
+          instance_double(Messages::AudioTranscriptionService, perform: { success: true, transcriptions: 'First audio text. ' })
+        )
+        allow(Messages::AudioTranscriptionService).to receive(:new).with(audio2).and_return(
+          instance_double(Messages::AudioTranscriptionService, perform: { success: true, transcriptions: 'Second audio text.' })
+        )
+      end
+
+      it 'concatenates all successful transcriptions' do
+        audio1 # trigger creation
+        audio2 # trigger creation
+
+        attachments = message.attachments
+        result = service.send(:extract_audio_transcriptions, attachments)
+        expect(result).to eq('First audio text. Second audio text.')
+      end
+    end
+
+    context 'with failed audio transcriptions' do
+      let(:audio_attachment) do
+        attachment = message.attachments.build(account_id: message.account_id, file_type: :audio)
+        attachment.save!
+        attachment
+      end
+
+      before do
+        allow(Messages::AudioTranscriptionService).to receive(:new).with(audio_attachment).and_return(
+          instance_double(Messages::AudioTranscriptionService, perform: { success: false, transcriptions: nil })
+        )
+      end
+
+      it 'returns empty string for failed transcriptions' do
+        audio_attachment # trigger creation
+
+        attachments = message.attachments
+        result = service.send(:extract_audio_transcriptions, attachments)
+        expect(result).to eq('')
+      end
+    end
+  end
+
+  describe 'private helper methods' do
+    describe '#text_part' do
+      it 'returns correct text part format' do
+        result = service.send(:text_part, 'Hello world')
+        expect(result).to eq({ type: 'text', text: 'Hello world' })
+      end
+    end
+
+    describe '#image_part' do
+      it 'returns correct image part format' do
+        result = service.send(:image_part, 'https://example.com/image.jpg')
+        expect(result).to eq({ type: 'image_url', image_url: { url: 'https://example.com/image.jpg' } })
+      end
+    end
+  end
+end

spec/enterprise/services/captain/tools/pdf_processing_service_spec.rb

@@ -0,0 +1,76 @@
+require 'rails_helper'
+
+RSpec.describe Captain::Tools::PdfProcessingService do
+  let(:account) { create(:account) }
+  let(:assistant) { create(:captain_assistant, account: account) }
+  let(:document) do
+    assistant.documents.create!(
+      name: 'Test PDF',
+      external_link: 'https://example.com/test.pdf',
+      document_type: :pdf
+    )
+  end
+
+  subject { described_class.new(document) }
+
+  describe '#process' do
+    context 'when document is not a PDF' do
+      before { allow(document).to receive(:pdf_document?).and_return(false) }
+
+      it 'returns early without processing' do
+        expect(subject.process).to be_nil
+      end
+    end
+
+    context 'when document is a PDF' do
+      let(:mock_file) { double('file', path: '/tmp/test.pdf', close: nil, unlink: nil) }
+      let(:mock_batcher) { double('batcher') }
+      let(:mock_vision) { double('vision') }
+
+      before do
+        allow(Down).to receive(:download).and_return(mock_file)
+        allow(Dir).to receive(:mktmpdir).and_return('/tmp/pdf_images')
+        allow(Dir).to receive(:glob).and_return(['/tmp/image1.png', '/tmp/image2.png'])
+        allow(FileUtils).to receive(:rm_rf)
+        
+        allow(Captain::Tools::PdfPageBatcherService).to receive(:new).and_return(mock_batcher)
+        allow(mock_batcher).to receive(:batch_pages).and_return(['/tmp/batched1.png'])
+        
+        allow(Captain::Tools::VisionTextExtractorService).to receive(:new).and_return(mock_vision)
+        allow(mock_vision).to receive(:extract_text_from_multiple_images).and_return('Extracted text content')
+      end
+
+      it 'processes the PDF and updates document' do
+        expect { subject.process }.to change { document.reload.status }.to('available')
+      end
+
+      it 'sets extracted content on the document' do
+        subject.process
+        expect(document.reload.content).to eq('Extracted text content')
+      end
+
+      it 'uses batching service for page processing' do
+        expect(mock_batcher).to receive(:batch_pages).with(['/tmp/image1.png', '/tmp/image2.png'])
+        subject.process
+      end
+
+      it 'uses Vision API for text extraction' do
+        expect(mock_vision).to receive(:extract_text_from_multiple_images).with(['/tmp/batched1.png'])
+        subject.process
+      end
+    end
+
+    context 'when processing fails' do
+      before do
+        allow(Down).to receive(:download).and_raise(StandardError.new('Download failed'))
+      end
+
+      it 'logs error and updates document with error message' do
+        expect(Rails.logger).to receive(:error).with(/PDF processing failed/)
+        
+        expect { subject.process }.to raise_error(StandardError)
+        expect(document.reload.content).to include('Error processing PDF')
+      end
+    end
+  end
+end

spec/enterprise/services/captain/tools/vision_text_extractor_service_spec.rb

@@ -0,0 +1,98 @@
+require 'rails_helper'
+
+RSpec.describe Captain::Tools::VisionTextExtractorService do
+  let(:image_path) { '/tmp/test_image.png' }
+
+  describe 'when Google Vision is available' do
+    subject { described_class.new }
+
+    let(:mock_vision) { double('vision_client') }
+    let(:mock_response) { double('response', error: nil) }
+    let(:mock_text_annotation) { double('text_annotation', text: 'Vision extracted text') }
+
+    before do
+      allow(Google::Cloud::Vision).to receive(:image_annotator).and_return(mock_vision)
+      allow(Google::Cloud::Vision::Image).to receive(:new).with(image_path).and_return(double('image'))
+      allow(mock_vision).to receive(:text_detection).and_return(mock_response)
+      allow(mock_response).to receive(:full_text_annotation).and_return(mock_text_annotation)
+    end
+
+    describe '#extract_text_from_image' do
+      it 'uses Vision API for text extraction' do
+        result = subject.extract_text_from_image(image_path)
+        expect(result).to eq('Vision extracted text')
+      end
+
+      context 'when Vision API returns error' do
+        before do
+          allow(mock_response).to receive(:error).and_return(double('error', message: 'API Error'))
+        end
+
+        it 'raises TextExtractionError' do
+          expect { subject.extract_text_from_image(image_path) }.to raise_error(
+            Captain::Tools::VisionTextExtractorService::TextExtractionError,
+            /Vision API error/
+          )
+        end
+      end
+    end
+  end
+
+  describe 'when Google Vision is not available' do
+    subject { described_class.new }
+
+    let(:mock_tesseract) { double('tesseract', to_s: 'Tesseract extracted text') }
+
+    before do
+      allow(Google::Cloud::Vision).to receive(:image_annotator).and_raise(StandardError.new('No credentials'))
+      allow(RTesseract).to receive(:new).with(image_path).and_return(mock_tesseract)
+    end
+
+    describe '#extract_text_from_image' do
+      it 'falls back to Tesseract for text extraction' do
+        result = subject.extract_text_from_image(image_path)
+        expect(result).to eq('Tesseract extracted text')
+      end
+
+      context 'when Tesseract fails' do
+        before do
+          allow(RTesseract).to receive(:new).and_raise(StandardError.new('Tesseract error'))
+        end
+
+        it 'raises TextExtractionError' do
+          expect { subject.extract_text_from_image(image_path) }.to raise_error(
+            Captain::Tools::VisionTextExtractorService::TextExtractionError,
+            /Failed to extract text with Tesseract/
+          )
+        end
+      end
+    end
+  end
+
+  describe '#extract_text_from_multiple_images' do
+    let(:image_paths) { ['/tmp/image1.png', '/tmp/image2.png'] }
+    
+    before do
+      allow(Google::Cloud::Vision).to receive(:image_annotator).and_return(double('vision'))
+      allow(subject).to receive(:extract_text_from_image).with('/tmp/image1.png').and_return('Text from page 1')
+      allow(subject).to receive(:extract_text_from_image).with('/tmp/image2.png').and_return('Text from page 2')
+    end
+
+    it 'extracts text from multiple images and joins them' do
+      result = subject.extract_text_from_multiple_images(image_paths)
+      expect(result).to eq("Text from page 1\n\nText from page 2")
+    end
+
+    it 'skips empty results' do
+      allow(subject).to receive(:extract_text_from_image).with('/tmp/image2.png').and_return('')
+      
+      result = subject.extract_text_from_multiple_images(image_paths)
+      expect(result).to eq('Text from page 1')
+    end
+
+    it 'logs which extraction method is being used' do
+      expect(Rails.logger).to receive(:info).with(/Using (Google Vision|Tesseract) for text extraction/)
+      subject.extract_text_from_multiple_images(image_paths)
+    end
+  end
+end