Class: Legion::LLM::API::ClientTranslators::OpenAIResponses::Events

Inherits:
Object
  • Object
show all
Includes:
Legion::Logging::Helper
Defined in:
lib/legion/llm/api/client_translators/openai_responses.rb

Overview

SSE Events emitter for /v1/responses. Emits sequence_number'd events following the OpenAI Responses API spec.

Instance Method Summary collapse

Constructor Details

#initialize(out:, request_id:, model:, conv_id: nil) ⇒ Events

Returns a new instance of Events.



249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 249

def initialize(out:, request_id:, model:, conv_id: nil)
  @out = out
  @request_id = request_id
  @model = model
  @conv_id = conv_id
  @seq = 0
  @output_items = []
  @msg_id = "msg_#{SecureRandom.hex(12)}"
  @msg_index = nil
  @msg_opened = false
  @full_text = +''
  @full_reasoning = +''
  @thinking_state = nil
  @pending_tool_calls = {} # block_index => { id, name, arguments_str, output_index }
  @created_at = Time.now.to_i
end

Instance Method Details

#on_done(stop_reason:, usage:, model:) ⇒ Object 



423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 423

def on_done(stop_reason:, usage:, model:)
  close_thinking_item

  if @msg_opened
    emit('response.output_text.done', {
           type: 'response.output_text.done', sequence_number: next_seq,
           output_index: @msg_index, content_index: 0, item_id: @msg_id, text: @full_text
         })
    emit('response.content_part.done', {
           type: 'response.content_part.done', sequence_number: next_seq,
           output_index: @msg_index, content_index: 0, item_id: @msg_id,
           part: { type: 'output_text', text: @full_text, annotations: [] }
         })
    completed_item = { id: @msg_id, type: 'message', role: 'assistant', status: 'completed',
                       content: [{ type: 'output_text', text: @full_text, annotations: [] }] }
    emit('response.output_item.done', {
           type: 'response.output_item.done', sequence_number: next_seq,
           output_index: @msg_index, item: completed_item
         })
    @output_items[@msg_index] = completed_item
  end

  # Responses protocol: every turn terminates with response.completed
  # (status completed). Client-callable function_call items ride in
  # output[]; the client executes them and continues via
  # function_call_output. requires_action / response.done are Assistants
  # API concepts that real Responses clients (Codex) reject — emitting
  # them closes the stream "before response.completed".
  _ = stop_reason
  emit('response.completed', {
         type: 'response.completed', sequence_number: next_seq,
         response: {
           id: @request_id, object: 'response', created_at: @created_at,
           status: 'completed', model: model.to_s,
           output: @output_items, usage: format_usage(usage)
         }
       })
end

#on_error(message:, type:, status_code:) ⇒ Object 



462
463
464
465
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 462

def on_error(message:, type:, status_code:)
  _ = status_code
  emit('error', { type: 'error', message: message, code: type })
end

#on_keep_alive ⇒ Object 



411
412
413
414
415
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 411

def on_keep_alive
  # SSE comment line — keeps the connection alive without issuing
  # a typed event.
  @out << ": keep-alive\n\n"
end

#on_message_delta(stop_reason:, output_tokens:) ⇒ Object 



417
418
419
420
421
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 417

def on_message_delta(stop_reason:, output_tokens:)
  _ = stop_reason
  _ = output_tokens
  # Consolidated trailer happens in on_done.
end

#on_server_tool_result(block_index:, tool_call_id:, result_text:) ⇒ Object 



402
403
404
405
406
407
408
409
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 402

def on_server_tool_result(block_index:, tool_call_id:, result_text:)
  _ = block_index
  _ = tool_call_id
  _ = result_text
  # /v1/responses doesn't expose server tool results inline today —
  # the function_call output item carries id/args; result is part
  # of the assistant message that follows on the next turn.
end

#on_start(model:, request_id:, input_tokens:) ⇒ Object 



266
267
268
269
270
271
272
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 266

def on_start(model:, request_id:, input_tokens:)
  _ = input_tokens
  base = { id: request_id, object: 'response', created_at: @created_at,
           status: 'in_progress', model: model.to_s, output: [], usage: nil }
  emit('response.created',     { type: 'response.created',     sequence_number: next_seq, response: base })
  emit('response.in_progress', { type: 'response.in_progress', sequence_number: next_seq, response: base })
end

#on_text_close(block_index:) ⇒ Object 



290
291
292
293
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 290

def on_text_close(block_index:)
  _ = block_index
  # Closure happens at on_done.
end

#on_text_delta(block_index:, text:) ⇒ Object 



279
280
281
282
283
284
285
286
287
288
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 279

def on_text_delta(block_index:, text:)
  _ = block_index
  close_thinking_item
  open_message
  @full_text << text
  emit('response.output_text.delta', {
         type: 'response.output_text.delta', sequence_number: next_seq,
         output_index: @msg_index, content_index: 0, item_id: @msg_id, delta: text
       })
end

#on_text_open(block_index:) ⇒ Object 



274
275
276
277
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 274

def on_text_open(block_index:)
  _ = block_index
  open_message
end

#on_thinking_close(block_index:, signature:) ⇒ Object 



333
334
335
336
337
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 333

def on_thinking_close(block_index:, signature:)
  _ = block_index
  _ = signature
  close_thinking_item
end

#on_thinking_delta(block_index:, text:, signature:) ⇒ Object 



319
320
321
322
323
324
325
326
327
328
329
330
331
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 319

def on_thinking_delta(block_index:, text:, signature:)
  _ = block_index
  _ = signature
  return if text.to_s.empty? || @thinking_state.nil?

  @full_reasoning << text
  @thinking_state[:thinking] << text
  output_index = @output_items.index(@thinking_state)
  emit('response.thinking.delta', {
         type: 'response.thinking.delta', sequence_number: next_seq,
         output_index: output_index, item_id: @thinking_state[:id], delta: text
       })
end

#on_thinking_open(block_index:) ⇒ Object 



295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 295

def on_thinking_open(block_index:)
  _ = block_index
  return if @thinking_state

  state = {
    type:     'thinking',
    id:       "thnk_#{SecureRandom.hex(12)}",
    thinking: +'',
    status:   'in_progress'
  }
  @output_items << state
  @thinking_state = state
  output_index = @output_items.length - 1
  emit('response.output_item.added', {
         type: 'response.output_item.added', sequence_number: next_seq,
         output_index: output_index, item: state
       })
  emit('response.thinking_part.added', {
         type: 'response.thinking_part.added', sequence_number: next_seq,
         output_index: output_index, item_id: state[:id],
         part: { type: 'thinking', thinking: '' }
       })
end

#on_tool_call_abort(block_index:, reason:) ⇒ Object

rubocop:disable Lint/UnusedMethodArgument



398
399
400
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 398

def on_tool_call_abort(block_index:, reason:) # rubocop:disable Lint/UnusedMethodArgument
  nil
end

#on_tool_call_close(block_index:) ⇒ Object 



380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 380

def on_tool_call_close(block_index:)
  state = @pending_tool_calls[block_index]
  return if state.nil?

  emit('response.function_call_arguments.done', {
         type: 'response.function_call_arguments.done', sequence_number: next_seq,
         output_index: state[:output_index], item_id: state[:id],
         arguments: state[:arguments_str]
       })
  completed = { id: state[:id], type: 'function_call', name: state[:name],
                call_id: state[:id], arguments: state[:arguments_str], status: 'completed' }
  emit('response.output_item.done', {
         type: 'response.output_item.done', sequence_number: next_seq,
         output_index: state[:output_index], item: completed
       })
  @output_items[state[:output_index]] = completed
end

#on_tool_call_delta(block_index:, partial_arguments_json:) ⇒ Object 



359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 359

def on_tool_call_delta(block_index:, partial_arguments_json:)
  state = @pending_tool_calls[block_index]
  return if state.nil?

  # If the assembler hands us a cumulative string (buffered close)
  # only emit the diff vs already emitted.
  new_part = if partial_arguments_json.start_with?(state[:args_emitted])
               partial_arguments_json[state[:args_emitted].length..]
             else
               partial_arguments_json
             end
  return if new_part.to_s.empty?

  state[:args_emitted] << new_part
  state[:arguments_str] = state[:args_emitted]
  emit('response.function_call_arguments.delta', {
         type: 'response.function_call_arguments.delta', sequence_number: next_seq,
         output_index: state[:output_index], item_id: state[:id], delta: new_part
       })
end

#on_tool_call_open(block_index:, tool_call:, server_tool:) ⇒ Object 



339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 339

def on_tool_call_open(block_index:, tool_call:, server_tool:)
  _ = server_tool
  tc_id = tool_call[:id] || "call_#{SecureRandom.hex(12)}"
  idx = @output_items.length
  item = { id: tc_id, type: 'function_call', name: tool_call[:name].to_s,
           call_id: tc_id, arguments: '', status: 'in_progress' }
  @output_items << item
  @pending_tool_calls[block_index] = {
    id:            tc_id,
    name:          tool_call[:name].to_s,
    arguments_str: '',
    output_index:  idx,
    args_emitted:  +''
  }
  emit('response.output_item.added', {
         type: 'response.output_item.added', sequence_number: next_seq,
         output_index: idx, item: item
       })
end