Module: Smith::Providers::OpenAI::Routing

Defined in:
lib/smith/providers/openai/routing.rb

Overview

Prepended onto RubyLLM::Providers::OpenAI. Intercepts the chat complete call and routes to /v1/responses when the rendered payload's openai_api_mode hint (set by the normalizer via chat.with_params(openai_api_mode: :responses)) requests it.

Does NOT rename RubyLLM::Providers::OpenAI::Chat#completion_url (PR #770 does; Smith diverges to keep the surface narrower). The instance_of? check prevents routing on OpenAI-compatible subclasses (OpenRouter, Azure, Bedrock).

The full /v1/responses payload assembly lives in Smith::Providers::OpenAI::Responses (vendored from crmne/ruby_llm PR #770 at SHA a84517db65d3774c6b129dc88032fe32c8dbc722). When the PR merges upstream, Smith bumps the ruby_llm dep and deletes the vendored files. The defined? guard in route_via_responses keeps the routing safe even if a host pins an older Smith without the vendored adapter, raising a clear NotImplementedError rather than silently falling through to chat-completions (which would still fail with the original tools+reasoning combo error).

Instance Method Summary collapse

Instance Method Details

#complete(messages, tools:, temperature:, model:, params: {}, headers: {}, schema: nil, thinking: nil, tool_prefs: nil) ⇒ Object 



29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
 
# File 'lib/smith/providers/openai/routing.rb', line 29

def complete(messages, tools:, temperature:, model:, params: {}, headers: {},
             schema: nil, thinking: nil, tool_prefs: nil, &)
  mode = params[:openai_api_mode] || params["openai_api_mode"]
  if mode.to_s == "responses" && instance_of?(::RubyLLM::Providers::OpenAI)
    route_via_responses(
      messages,
      tools: tools, temperature: temperature, model: model,
      params: params.except(:openai_api_mode, "openai_api_mode"),
      headers: headers, schema: schema, thinking: thinking,
      tool_prefs: tool_prefs, &
    )
  else
    super
  end
end