RubyLLM Adapter

Bridge RubyLLM chat completion events into the OpenUI Rails stream contract.

After reading this guide, you will know:

  • How to pass a RubyLLM chat through the optional adapter.
  • How the adapter applies the generated library instruction and returns envelopes.
  • Which RubyLLM version this integration has been verified against.

Stream a RubyLLM chat

RubyLLM is optional. Add it to the host app separately, configure its provider credentials through RubyLLM, and build a chat using the host’s normal RubyLLM setup.

transport = OpenUI::Rails::Stream::Transports.build(
  :action_cable,
  stream_name: "openui:surface:#{surface.id}"
)

OpenUI::Rails::Adapters::RubyLLM.stream(
  chat: chat,
  question: "Summarize this ticket",
  surface_id: surface.id
) do |envelope|
  transport.broadcast(envelope)
end

The default :complete strategy adds the question as a user message and calls complete with a block. The adapter loads the system instruction from the configured prompt and manifest, passes it through with_instructions when available, and normalizes string chunks or chunks with content. Choose strategy: :ask if the host uses RubyLLM’s ask API.

The adapter uses the configured byte limits and emits ordered Stream::Envelope objects. It converts provider failures into structured error frames without including the provider’s error text. Exceptions raised by your envelope consumer propagate to the host.

Compatibility boundary

The repository verifies RubyLLM 2.1.1 with deterministic OpenAI chat-completions SSE. Other versions and providers are not guaranteed by that check. The adapter does not make RubyLLM a gem dependency and does not configure provider credentials.

Next Steps