ruby_llm 2.0.0.rc2 → 2.0.0.rc4

data/lib/ruby_llm/moderation.rb

@@ -104,7 +104,7 @@ module RubyLLM
       empty_tokens = Tokens.new
       payload = {
         provider: provider_instance.slug,
-        provider_class: provider_instance.class.display_name,
+        provider_class: provider_instance.name,
         model: model&.id,
         model_info: model,
         input: input,

data/lib/ruby_llm/ocr.rb

@@ -62,7 +62,7 @@ module RubyLLM
                                                        config: config)
       payload = {
         provider: provider_instance.slug,
-        provider_class: provider_instance.class.display_name,
+        provider_class: provider_instance.name,
         model: model.id,
         model_info: model,
         pages: pages,

data/lib/ruby_llm/protocols/anthropic/chat.rb

@@ -39,13 +39,12 @@ module RubyLLM
           warn_unsupported_citations(model) if citations && !model.supports?(:citations)
           tool_prefs ||= {}
           system_messages, chat_messages = separate_messages(messages)
-          explicit_boundaries = cache_boundaries?(messages, caching:)
           system_content = build_system_content(system_messages, caching:)
 
           build_base_payload(chat_messages, model, stream, thinking, citations: citations, caching:,
                                                                      max_output_tokens:).tap do |payload|
             add_optional_fields(payload, system_content:, tools:, tool_prefs:, temperature:, schema:)
-            payload[:cache_control] = prompt_cache_control(caching) if caching && !explicit_boundaries
+            payload[:cache_control] = prompt_cache_control(caching) if caching
           end
         end
 
@@ -429,10 +428,6 @@ module RubyLLM
           content_blocks.concat(Media.format_content(msg.content, msg.attachments, citations: citations))
         end
 
-        def cache_boundaries?(messages, caching: nil)
-          caching != false && messages.any?(&:cache_until_here?)
-        end
-
         def cache_boundary?(message, caching: nil)
           caching != false && message.cache_until_here?
         end

data/lib/ruby_llm/protocols/bedrock/async_videos.rb

@@ -64,7 +64,8 @@ module RubyLLM
           prefix = job.raw.dig('outputDataConfig', 's3OutputDataConfig', 's3Uri')
           raise Error, 'Bedrock video job has no output S3 URI' unless prefix
 
-          uris = @provider.list_file_uris("#{prefix.sub(%r{/+\z}, '')}/")
+          prefix_length = (prefix.rindex(%r{[^/]}) || -1) + 1
+          uris = @provider.list_file_uris("#{prefix[0, prefix_length]}/")
           videos = uris.select { |uri| uri.downcase.end_with?('.mp4') }
           raise Error, 'Expected exactly one MP4 in the Bedrock video output' unless videos.one?
 

data/lib/ruby_llm/protocols/chat_completions/chat.rb

@@ -87,7 +87,7 @@ module RubyLLM
           payload[:reasoning_effort] = effort if effort
 
           payload[:stream_options] = { include_usage: true } if stream
-          apply_prompt_cache_params(payload, messages, caching)
+          apply_prompt_cache_params(payload, caching)
           payload
         end
 
@@ -250,11 +250,10 @@ module RubyLLM
           end
         end
 
-        def apply_prompt_cache_params(payload, messages, caching)
+        def apply_prompt_cache_params(payload, caching)
           return unless openai_prompt_caching?
 
           payload.merge!(prompt_cache_params(caching)) if caching
-          force_explicit_cache_mode(payload) if caching != false && cache_boundaries?(messages)
         end
 
         def openai_prompt_caching?
@@ -290,14 +289,6 @@ module RubyLLM
           retention
         end
 
-        def force_explicit_cache_mode(payload)
-          payload[:prompt_cache_options] = { mode: 'explicit' }.merge(payload[:prompt_cache_options] || {})
-        end
-
-        def cache_boundaries?(messages)
-          messages.any?(&:cache_until_here?)
-        end
-
         def prompt_cache_options(caching)
           options = caching.to_h.transform_keys(&:to_sym)
           unsupported = options.keys - PROMPT_CACHE_OPTIONS

data/lib/ruby_llm/protocols/chat_completions/rerank.rb

@@ -38,10 +38,13 @@ module RubyLLM
         def parse_rerank_results(data, documents = [])
           Array(data['results']).map do |result|
             index = result['index']
+            unless valid_rerank_index?(index, documents)
+              raise Error, 'Rerank endpoint returned an invalid document index'
+            end
 
             RubyLLM::Rerank::Result.new(
               index: index,
-              document: rerank_document(result['document']) || (index && documents[index]),
+              document: rerank_document(result['document']) || documents[index],
               score: result['relevance_score']
             )
           end
@@ -50,6 +53,10 @@ module RubyLLM
         def rerank_document(document)
           document.is_a?(Hash) ? document['text'] : document
         end
+
+        def valid_rerank_index?(index, documents)
+          index.is_a?(Integer) && index.between?(0, documents.length - 1)
+        end
       end
     end
   end

data/lib/ruby_llm/protocols/cohere/rerank.rb

@@ -38,14 +38,21 @@ module RubyLLM
         def parse_rerank_results(data, documents)
           Array(data['results']).map do |result|
             index = result['index']
+            unless valid_rerank_index?(index, documents)
+              raise Error, 'Cohere reranking returned an invalid document index'
+            end
 
             RubyLLM::Rerank::Result.new(
               index: index,
-              document: result.dig('document', 'text') || (index && documents[index]),
+              document: result.dig('document', 'text') || documents[index],
               score: result['relevance_score']
             )
           end
         end
+
+        def valid_rerank_index?(index, documents)
+          index.is_a?(Integer) && index.between?(0, documents.length - 1)
+        end
       end
     end
   end

data/lib/ruby_llm/protocols/converse/chat.rb

@@ -266,7 +266,6 @@ module RubyLLM
 
         def automatic_cache_target(system_messages, chat_messages, caching)
           return unless caching
-          return if (system_messages + chat_messages).any?(&:cache_until_here?)
 
           (chat_messages.reverse + system_messages.reverse).find { |msg| cacheable_message?(msg) }
         end

data/lib/ruby_llm/protocols/gemini/embedding_batches.rb

@@ -60,6 +60,11 @@ module RubyLLM
           metadata = responses.first.fetch('metadata')
           return if responses.size < metadata.fetch('embedding_count')
 
+          positions = responses.map { |inline| inline.dig('metadata', 'embedding_index') }
+          unless positions.sort == (0...responses.size).to_a
+            return [index, nil, batch_failure(key, 'Invalid or duplicate embedding record positions')]
+          end
+
           vectors = embedding_batch_vectors(responses)
           return [index, nil, batch_failure(key, 'Gemini returned no embedding')] unless vectors
 

data/lib/ruby_llm/protocols/interactions/tools.rb

@@ -31,7 +31,9 @@ module RubyLLM
         end
 
         def parse_interaction_arguments(arguments)
-          arguments.is_a?(String) ? JSON.parse(arguments) : arguments || {}
+          return {} if arguments.nil? || (arguments.is_a?(String) && arguments.empty?)
+
+          arguments.is_a?(String) ? JSON.parse(arguments) : arguments
         rescue JSON::ParserError => e
           raise ToolCallParseError.new(finish_reason: :tool_calls), cause: e
         end

data/lib/ruby_llm/protocols/responses/chat.rb

@@ -12,7 +12,6 @@ module RubyLLM
         OPENAI_INLINE_FILE_LIMIT = 50 * 1024 * 1024
         OPENAI_FILE_UPLOAD_LIMIT = 512 * 1024 * 1024
         PROMPT_CACHE_OPTIONS = %i[key ttl mode retention].freeze
-        CACHE_BREAKPOINT_ROLES = %i[user system].freeze
 
         module_function
 
@@ -47,7 +46,6 @@ module RubyLLM
           payload[:reasoning] = { effort: effort } if effort
           payload[:reasoning] = (payload[:reasoning] || {}).merge(summary: 'auto') if thinking&.display == :summarized
           payload.merge!(prompt_cache_params(caching)) if caching
-          force_explicit_cache_mode(payload) if caching != false && cache_boundaries?(messages)
 
           payload
         end
@@ -225,14 +223,6 @@ module RubyLLM
           retention
         end
 
-        def force_explicit_cache_mode(payload)
-          payload[:prompt_cache_options] = { mode: 'explicit' }.merge(payload[:prompt_cache_options] || {})
-        end
-
-        def cache_boundaries?(messages)
-          messages.any? { |msg| msg.cache_until_here? && CACHE_BREAKPOINT_ROLES.include?(msg.role) }
-        end
-
         def prompt_cache_options(caching)
           options = caching.to_h.transform_keys(&:to_sym)
           unsupported = options.keys - PROMPT_CACHE_OPTIONS

data/lib/ruby_llm/providers/deepseek/responses.rb

@@ -8,7 +8,6 @@ module RubyLLM
       # instead of summaries, both in responses and in the stream.
       class Responses < Protocols::Responses
         SERVER_TOOL_ALIASES = {
-          web_search: { tool: { type: 'web_search' } },
           apply_patch: { tool: { type: 'custom', name: 'apply_patch' } }
         }.freeze
 

data/lib/ruby_llm/providers/mistral/ocr.rb

@@ -22,7 +22,7 @@ module RubyLLM
         end
 
         def ocr_document_part(attachment)
-          reference = attachment.url? ? attachment.source.to_s : attachment.for_llm
+          reference = attachment.url? ? attachment.source.to_s : ocr_data_uri(attachment)
 
           if attachment.image?
             { type: 'image_url', image_url: reference }
@@ -31,6 +31,10 @@ module RubyLLM
           end
         end
 
+        def ocr_data_uri(attachment)
+          "data:#{attachment.mime_type};base64,#{attachment.encoded}"
+        end
+
         def parse_ocr_response(response, model:)
           data = response.body
 

data/lib/ruby_llm/providers/openrouter/chat.rb

@@ -47,7 +47,7 @@ module RubyLLM
 
           reasoning = build_reasoning(thinking)
           payload[:reasoning] = reasoning if reasoning
-          payload[:cache_control] = prompt_cache_control(caching) if caching && !cache_boundaries?(messages)
+          payload[:cache_control] = prompt_cache_control(caching) if caching
           payload
         end
 
@@ -142,10 +142,6 @@ module RubyLLM
           keys.map { |key| ":#{key}" }.join(', ')
         end
 
-        def cache_boundaries?(messages)
-          messages.any?(&:cache_until_here?)
-        end
-
         def openai_prompt_caching?
           false
         end

data/lib/ruby_llm/rerank.rb

@@ -81,7 +81,7 @@ module RubyLLM
       empty_tokens = Tokens.new
       payload = {
         provider: provider_instance.slug,
-        provider_class: provider_instance.class.display_name,
+        provider_class: provider_instance.name,
         model: model.id,
         model_info: model,
         query: query,

data/lib/ruby_llm/speech.rb

@@ -91,7 +91,7 @@ module RubyLLM
 
       payload = {
         provider: provider_instance.slug,
-        provider_class: provider_instance.class.display_name,
+        provider_class: provider_instance.name,
         model: model.id,
         model_info: model,
         input: input,

data/lib/ruby_llm/transcription.rb

@@ -132,7 +132,7 @@ module RubyLLM
       empty_tokens = Tokens.new
       payload = {
         provider: provider_instance.slug,
-        provider_class: provider_instance.class.display_name,
+        provider_class: provider_instance.name,
         model: model.id,
         model_info: model,
         language: language,

data/lib/ruby_llm/version.rb

@@ -2,5 +2,5 @@
 
 module RubyLLM
   # The version of the ruby_llm gem, as a string.
-  VERSION = '2.0.0.rc2'
+  VERSION = '2.0.0.rc4'
 end

data/lib/ruby_llm/video_job.rb

@@ -50,7 +50,7 @@ module RubyLLM
                                                        config: config)
       payload = {
         provider: provider_instance.slug,
-        provider_class: provider_instance.class.display_name,
+        provider_class: provider_instance.name,
         model: model.id,
         model_info: model,
         prompt: prompt,

data/lib/ruby_llm.rb

@@ -44,7 +44,7 @@ loader.setup
 
 # RubyLLM is an AI framework for Ruby and Rails. Build conversations and
 # agents, generate media, process documents, and work with model providers
-# through one Ruby API. The guides at https://rubyllm.com/next/ introduce
+# through one Ruby API. The guides at https://rubyllm.com/ introduce
 # each feature; this reference documents its classes, arguments, and results.
 #
 #   RubyLLM.configure do |config|

data/lib/tasks/ruby_llm.rake

@@ -7,7 +7,7 @@ namespace :ruby_llm do
       task action => :environment do
         require 'generators/ruby_llm/upgrade/upgrade_migration'
 
-        RubyLLM::Generators::UpgradeMigration.new.public_send(action)
+        RubyLLM::Generators::UpgradeMigration.for.public_send(action)
         puts "RubyLLM copy upgrade: #{action} completed"
       end
     end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: ruby_llm
 version: !ruby/object:Gem::Version
-  version: 2.0.0.rc2
+  version: 2.0.0.rc4
 platform: ruby
 authors:
 - Carmine Paolino
@@ -275,6 +275,11 @@ files:
 - lib/generators/ruby_llm/tool/templates/tool_call.html.erb.tt
 - lib/generators/ruby_llm/tool/templates/tool_result.html.erb.tt
 - lib/generators/ruby_llm/tool/tool_generator.rb
+- lib/generators/ruby_llm/upgrade/legacy_content_sql.rb
+- lib/generators/ruby_llm/upgrade/online_copy_migration.rb
+- lib/generators/ruby_llm/upgrade/online_copy_migration/data.rb
+- lib/generators/ruby_llm/upgrade/online_copy_migration/journal.rb
+- lib/generators/ruby_llm/upgrade/online_copy_migration/verification.rb
 - lib/generators/ruby_llm/upgrade/templates/backfill_v2_data.rb.tt
 - lib/generators/ruby_llm/upgrade/templates/cleanup_v2_upgrade.rb.tt
 - lib/generators/ruby_llm/upgrade/templates/finish_v2_upgrade.rb.tt
@@ -607,24 +612,24 @@ metadata:
   homepage_uri: https://rubyllm.com
   source_code_uri: https://github.com/crmne/ruby_llm
   changelog_uri: https://github.com/crmne/ruby_llm/releases
-  documentation_uri: https://rubyllm.com/next/
+  documentation_uri: https://rubyllm.com/
   bug_tracker_uri: https://github.com/crmne/ruby_llm/issues
   funding_uri: https://github.com/sponsors/crmne
   rubygems_mfa_required: 'true'
 post_install_message: |
-  RubyLLM 2.0.0.rc2
+  RubyLLM 2.0.0.rc4
 
     2.0 renames several APIs and changes what message content returns. Coming
     from 1.x? Read the upgrade guide before you boot:
 
-      https://rubyllm.com/next/upgrading/
+      https://rubyllm.com/upgrading/
 
     The Rails upgrade uses forward-only preparation, backfill, and finish
     phases, with cleanup later. Rename mode is the default; optional copy
     mode supports a controlled return to 1.16. Read its requirements in the
     upgrade guide and rehearse on a recent production snapshot.
 
-    Agent skill: https://rubyllm.com/next/ai-coding-assistants/
+    Agent skill: https://rubyllm.com/ai-coding-assistants/
 rdoc_options: []
 require_paths:
 - lib
@@ -639,7 +644,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 4.0.16
+rubygems_version: 4.0.20
 specification_version: 4
 summary: 'Build AI features the Ruby way: a delightful Ruby AI framework for every
   major AI provider.'