What this does

Adds two complementary features to RubyLLM:

1. prompt() method - Non-persisting chat interactions

A new method that works like ask() but doesn't persist anything to the database. Perfect for:

• A/B testing: Generate multiple responses with different temperatures/models, let users pick one
• Injecting: prompts without polluting chat history (tool calls side effects allowed intentionally)
• Speculative generation: Try different prompts/settings without commitment - generative variation without altering history

A/B testing example

msg = chat.messages.create!(role: :user, content: "Explain 'optionality'")
response_a = chat.with_temperature(0.3).prompt(msg)
response_b = chat.with_temperature(0.9).prompt(msg)

Then add create an assistant message with both responses and a 'choose' action. When user picks best answer, then we manually persist the winner (with backfilled tool calls if desired) removing the A/B selection content, replacing it with the selected response.

2. include_attachments: parameter - Performance optimization

Added optional parameter to Message#to_llm() to skip attachment downloads:

Window attachments: apply arbitrary logic to which messages ought have their attachments downloaded for each chat run

class Message
  # Overriding in the application 
  def to_llm(include_attachments: false)
    super(include_attachments: (include_attachments || created_at > 5.minutes.ago))
  end
end

Why this matters:

• Long chats with many attachments = performance issues
• Enables smart attachment windowing strategies
• Defaults to true (backward compatible)

Type of change

• New feature
• Bug fix
• Breaking change
• Documentation
• Performance improvement (attachment windowing)

Scope check

• I read the Contributing Guide
• This aligns with RubyLLM's focus on LLM communication
• This isn't application-specific logic that belongs in user code
• This benefits most users, not just my specific use case

Why this belongs in RubyLLM:

• prompt() is a fundamental LLM interaction pattern (like ask() but non-persisting)
• Attachment windowing is a general performance concern for multimodal chats
• Both features enable common patterns: A/B testing, attachment windowing

Quality check

• I ran overcommit --install and all hooks pass
• I tested my changes thoroughly
  • For provider changes: Re-recorded VCR cassettes with bundle exec rake vcr:record[provider_name] - Requires maintainer with API credentials
  • All tests pass
• I updated documentation if needed - Would appreciate guidance on where docs should go
• I didn't modify auto-generated files manually (models.json, aliases.json)

Test Coverage:

• Added tests for prompt() method
• Added tests for include_attachments: parameter (all passing)

API changes

• New public methods/classes
  • Chat#prompt(message, with: nil, &block) (both new and legacy APIs)
  • Message#prompt_messages (singleton method on response) for accessibility
• Changed method signatures
  • Message#to_llm(include_attachments: true) (optional parameter, defaults to true)
• Breaking change
• No API changes

Backward Compatibility:

• ✅ include_attachments: defaults to true (existing behavior unchanged)
• ✅ prompt() is additive
• ✅ Feature parity: Both features added to new AND legacy APIs (respects use_new_acts_as config)

Related issues

Not sure.