Module: Legion::LLM::API::OpenAI::Responses

Extended by:
Legion::Logging::Helper
Defined in:
lib/legion/llm/api/openai/responses.rb

Class Method Summary collapse

Class Method Details

.build_handler ⇒ Object 



25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
 
# File 'lib/legion/llm/api/openai/responses.rb', line 25

def self.build_handler
  proc do
    require_llm!
    body = parse_request_body
    request_id = "resp_#{SecureRandom.hex(16)}"

    input = body[:input]
    messages = case input
               when Array
                 Responses.normalize_input_array(input)
               when String
                 [{ role: 'user', content: input }]
               else
                 halt 400, { 'Content-Type' => 'application/json' },
                      Legion::JSON.dump({ error: { message: 'input is required (string or array)',
                                                   type: 'invalid_request_error', code: nil } })
               end

    messages = [{ role: 'system', content: body[:instructions].to_s }] + messages if body[:instructions]

    model = body[:model] || Legion::Settings[:llm][:default_model] || 'default'
    streaming = body[:stream] == true

    tool_declarations = Responses.build_tool_declarations(body[:tools])

    ext_provider = env['HTTP_X_LEGION_PROVIDER'] || body[:provider]
    ext_tier     = env['HTTP_X_LEGION_TIER']     || body[:tier]
    ext_instance = env['HTTP_X_LEGION_INSTANCE'] || body[:instance]

    routing = { provider: ext_provider, instance: ext_instance, model: model }.compact
    extra   = {}
    extra[:tier] = ext_tier.to_sym if ext_tier

    log.info(
      "[llm][api][openai][responses] action=accepted request_id=#{request_id} " \
      "model=#{model} stream=#{streaming} tools=#{tool_declarations.size}"
    )

    effective_caller = build_server_caller(source: 'openai_responses', path: request.path, env: env)

    require 'legion/llm/inference/request' unless defined?(Legion::LLM::Inference::Request)
    require 'legion/llm/inference/executor' unless defined?(Legion::LLM::Inference::Executor)

    inference_request = Legion::LLM::Inference::Request.build(
      id:       request_id,
      messages: messages,
      routing:  routing,
      tools:    tool_declarations,
      caller:   effective_caller,
      stream:   streaming,
      cache:    { strategy: :default, cacheable: true },
      extra:    extra.empty? ? {} : extra
    )

    executor = Legion::LLM::Inference::Executor.new(inference_request)

    if streaming
      content_type 'text/event-stream'
      headers 'Cache-Control'     => 'no-cache',
              'Connection'        => 'keep-alive',
              'X-Accel-Buffering' => 'no'

      stream do |out|
        Responses.stream_response(out, executor, request_id: request_id, model: model, upstream_body: body)
      rescue StandardError => e
        handle_exception(e, level: :error, handled: false, operation: 'llm.api.openai.responses.stream', request_id: request_id)
        out << "event: error\ndata: #{Legion::JSON.dump({ type: 'server_error', message: e.message })}\n\n"
      end
    else
      pipeline_response = executor.call_responses(body: body, stream: false)
      response_body = Responses.format_response(pipeline_response, request_id: request_id, model: model)

      log.info("[llm][api][openai][responses] action=complete request_id=#{request_id} model=#{response_body[:model]}")
      content_type :json
      status 200
      Legion::JSON.dump(response_body)
    end
  rescue Legion::LLM::AuthError => e
    handle_exception(e, level: :error, handled: true, operation: 'llm.api.openai.responses.auth')
    halt 401, { 'Content-Type' => 'application/json' },
         Legion::JSON.dump({ error: { message: e.message, type: 'authentication_error' } })
  rescue Legion::LLM::RateLimitError => e
    handle_exception(e, level: :warn, handled: true, operation: 'llm.api.openai.responses.rate_limit')
    halt 429, { 'Content-Type' => 'application/json' },
         Legion::JSON.dump({ error: { message: e.message, type: 'rate_limit_error' } })
  rescue Legion::LLM::ProviderDown, Legion::LLM::ProviderError => e
    handle_exception(e, level: :error, handled: true, operation: 'llm.api.openai.responses.provider')
    halt 502, { 'Content-Type' => 'application/json' },
         Legion::JSON.dump({ error: { message: e.message, type: 'server_error' } })
  rescue StandardError => e
    handle_exception(e, level: :error, handled: false, operation: 'llm.api.openai.responses')
    halt 500, { 'Content-Type' => 'application/json' },
         Legion::JSON.dump({ error: { message: e.message, type: 'server_error' } })
  end
end

.build_output_tool_calls(pipeline_response) ⇒ Object 



369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
 
# File 'lib/legion/llm/api/openai/responses.rb', line 369

def self.build_output_tool_calls(pipeline_response)
  tools_data = pipeline_response.respond_to?(:tools) ? pipeline_response.tools : nil
  return [] unless tools_data.is_a?(Array) && !tools_data.empty?

  tools_data.filter_map do |tc|
    name = tc.respond_to?(:name) ? tc.name : (tc[:name] || tc['name'])
    args = tc.respond_to?(:arguments) ? tc.arguments : (tc[:arguments] || tc['arguments'] || {})
    tc_id = tc.respond_to?(:id) ? tc.id : (tc[:id] || tc['id'] || "call_#{SecureRandom.hex(8)}")
    next unless name

    {
      type:      'function_call',
      id:        "fc_#{SecureRandom.hex(12)}",
      call_id:   tc_id,
      name:      name.to_s,
      arguments: args.is_a?(String) ? args : Legion::JSON.dump(args),
      status:    'completed'
    }
  end
end

.build_tool_declarations(tools) ⇒ Object 



166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
 
# File 'lib/legion/llm/api/openai/responses.rb', line 166

def self.build_tool_declarations(tools)
  return [] if tools.nil? || !tools.is_a?(Array) || tools.empty?

  tools.filter_map do |tool|
    t = tool.respond_to?(:transform_keys) ? tool.transform_keys(&:to_sym) : tool
    fn = t[:function] || t
    fn = fn.transform_keys(&:to_sym) if fn.respond_to?(:transform_keys)
    next unless fn[:name].to_s.length.positive?

    Legion::LLM::Types::ToolDefinition.build(
      name:        fn[:name].to_s,
      description: fn[:description].to_s,
      parameters:  fn[:parameters] || {},
      source:      { type: :client, executable: true }
    )
  rescue StandardError => e
    handle_exception(e, level: :warn, handled: true, operation: 'llm.api.openai.responses.build_tool')
    nil
  end
end

.build_usage(tokens) ⇒ Object 



413
414
415
416
417
418
419
420
421
422
 
# File 'lib/legion/llm/api/openai/responses.rb', line 413

def self.build_usage(tokens)
  input_tokens = extract_token(tokens, :input_tokens)
  output_tokens = extract_token(tokens, :output_tokens)

  {
    input_tokens:  input_tokens,
    output_tokens: output_tokens,
    total_tokens:  input_tokens + output_tokens
  }
end

.call_streaming_executor(executor, upstream_body: nil) ⇒ Object 



357
358
359
360
361
362
363
 
# File 'lib/legion/llm/api/openai/responses.rb', line 357

def self.call_streaming_executor(executor, upstream_body: nil, &)
  if upstream_body && executor.respond_to?(:call_responses)
    executor.call_responses(body: upstream_body, stream: true, &)
  else
    executor.call_stream(&)
  end
end

.extract_token(tokens, key) ⇒ Object 



390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
 
# File 'lib/legion/llm/api/openai/responses.rb', line 390

def self.extract_token(tokens, key)
  return 0 if tokens.nil?

  aliases = token_aliases(key)

  if tokens.is_a?(Hash)
    aliases.each do |candidate|
      value = tokens[candidate]
      value = tokens[candidate.to_s] if value.nil?
      return value.to_i unless value.nil?
    end

    return 0
  end

  aliases.each do |candidate|
    method_name = token_method(candidate)
    return tokens.public_send(method_name).to_i if method_name && tokens.respond_to?(method_name)
  end

  0
end

.flush_pending_tool_calls(messages, pending) ⇒ Object 



153
154
155
156
157
158
159
160
161
162
163
164
 
# File 'lib/legion/llm/api/openai/responses.rb', line 153

def self.flush_pending_tool_calls(messages, pending)
  return if pending.empty?

  messages << {
    role:       'assistant',
    content:    '',
    tool_calls: pending.map do |tc|
      { id: tc[:id], type: 'function', function: { name: tc[:name], arguments: tc[:arguments] } }
    end
  }
  pending.clear
end

.format_response(pipeline_response, request_id:, model:) ⇒ Object 



187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
 
# File 'lib/legion/llm/api/openai/responses.rb', line 187

def self.format_response(pipeline_response, request_id:, model:)
  routing = pipeline_response.routing || {}
  tokens = pipeline_response.tokens || {}
  raw_msg = pipeline_response.message
  content = raw_msg.is_a?(Hash) ? (raw_msg[:content] || raw_msg['content']).to_s : raw_msg.to_s
  resolved_model = (routing[:model] || routing['model'] || model).to_s

  output = []

  tool_calls = build_output_tool_calls(pipeline_response)
  output.concat(tool_calls)

  output << {
    type:    'message',
    id:      "msg_#{SecureRandom.hex(12)}",
    role:    'assistant',
    content: [{ type: 'output_text', text: content }],
    status:  'completed'
  }

  {
    id:         request_id,
    object:     'response',
    created_at: Time.now.to_i,
    model:      resolved_model,
    output:     output,
    usage:      build_usage(tokens),
    status:     'completed'
  }
end

.normalize_input_array(input) ⇒ Object 



121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
 
# File 'lib/legion/llm/api/openai/responses.rb', line 121

def self.normalize_input_array(input)
  messages = []
  pending_tool_calls = []

  input.each do |item|
    item = item.transform_keys(&:to_sym) if item.respond_to?(:transform_keys)
    case item[:type]&.to_s
    when 'function_call'
      pending_tool_calls << {
        id:        item[:call_id] || item[:id],
        name:      item[:name].to_s,
        arguments: item[:arguments].is_a?(String) ? item[:arguments] : Legion::JSON.dump(item[:arguments] || {})
      }
    when 'function_call_output'
      flush_pending_tool_calls(messages, pending_tool_calls)
      messages << { role: 'tool', tool_call_id: item[:call_id], content: item[:output].to_s }
    else
      flush_pending_tool_calls(messages, pending_tool_calls)
      role = item[:role]&.to_s
      next unless role

      role = 'system' if role == 'developer'

      content = item[:content]
      content = content.to_s if content && !content.is_a?(Array)
      messages << { role: role, content: content }.compact
    end
  end
  flush_pending_tool_calls(messages, pending_tool_calls)
  messages
end

.registered(app) ⇒ Object 



14
15
16
17
18
19
20
21
22
23
 
# File 'lib/legion/llm/api/openai/responses.rb', line 14

def self.registered(app)
  log.debug('[llm][api][openai][responses] registering POST /v1/responses + /api/llm/inference/v1/responses')

  handler = build_handler

  app.post('/v1/responses') { instance_exec(&handler) }
  app.post('/api/llm/inference/v1/responses') { instance_exec(&handler) }

  log.debug('[llm][api][openai][responses] routes registered')
end

.sse_event(name, payload) ⇒ Object 



365
366
367
 
# File 'lib/legion/llm/api/openai/responses.rb', line 365

def self.sse_event(name, payload)
  "event: #{name}\ndata: #{Legion::JSON.dump(payload)}\n\n"
end

.stream_response(out, executor, request_id:, model:, upstream_body: nil) ⇒ Object 



218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
 
# File 'lib/legion/llm/api/openai/responses.rb', line 218

def self.stream_response(out, executor, request_id:, model:, upstream_body: nil)
  created_at = Time.now.to_i
  seq = 0
  in_progress_response = { id: request_id, object: 'response', created_at: created_at,
                           status: 'in_progress', model: model, output: [], usage: nil }

  # response.created — envelope matches gateway format: { type:, response:, sequence_number: }
  out << sse_event('response.created', {
                     type:            'response.created',
                     sequence_number: seq += 1,
                     response:        in_progress_response
                   })

  out << sse_event('response.in_progress', {
                     type:            'response.in_progress',
                     sequence_number: seq += 1,
                     response:        in_progress_response
                   })

  msg_id = "msg_#{SecureRandom.hex(12)}"
  out << sse_event('response.output_item.added', {
                     type:            'response.output_item.added',
                     sequence_number: seq += 1,
                     output_index:    0,
                     item:            { id: msg_id, type: 'message', role: 'assistant',
                       content: [], status: 'in_progress' }
                   })

  out << sse_event('response.content_part.added', {
                     type:            'response.content_part.added',
                     sequence_number: seq += 1,
                     output_index:    0,
                     content_index:   0,
                     item_id:         msg_id,
                     part:            { type: 'output_text', text: '', annotations: [] }
                   })

  full_text = +''

  pipeline_response = call_streaming_executor(executor, upstream_body: upstream_body) do |chunk|
    text = chunk.respond_to?(:content) ? chunk.content.to_s : chunk.to_s
    next if text.empty?

    full_text << text
    out << sse_event('response.output_text.delta', {
                       type:            'response.output_text.delta',
                       sequence_number: seq += 1,
                       output_index:    0,
                       content_index:   0,
                       item_id:         msg_id,
                       delta:           text
                     })
  end

  routing = pipeline_response.routing || {}
  tokens  = pipeline_response.tokens || {}
  resolved_model = (routing[:model] || routing['model'] || model).to_s
  usage = build_usage(tokens)
  function_calls = build_output_tool_calls(pipeline_response)

  out << sse_event('response.output_text.done', {
                     type:            'response.output_text.done',
                     sequence_number: seq += 1,
                     output_index:    0,
                     content_index:   0,
                     item_id:         msg_id,
                     text:            full_text
                   })

  out << sse_event('response.content_part.done', {
                     type:            'response.content_part.done',
                     sequence_number: seq += 1,
                     output_index:    0,
                     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: [] }] }
  out << sse_event('response.output_item.done', {
                     type:            'response.output_item.done',
                     sequence_number: seq += 1,
                     output_index:    0,
                     item:            completed_item
                   })

  function_calls.each_with_index do |function_call, index|
    output_index = index + 1
    in_progress_item = function_call.merge(status: 'in_progress', arguments: '')

    out << sse_event('response.output_item.added', {
                       type:            'response.output_item.added',
                       sequence_number: seq += 1,
                       output_index:    output_index,
                       item:            in_progress_item
                     })

    out << sse_event('response.function_call_arguments.delta', {
                       type:            'response.function_call_arguments.delta',
                       sequence_number: seq += 1,
                       output_index:    output_index,
                       item_id:         function_call[:id],
                       delta:           function_call[:arguments]
                     })

    out << sse_event('response.function_call_arguments.done', {
                       type:            'response.function_call_arguments.done',
                       sequence_number: seq += 1,
                       output_index:    output_index,
                       item_id:         function_call[:id],
                       arguments:       function_call[:arguments]
                     })

    out << sse_event('response.output_item.done', {
                       type:            'response.output_item.done',
                       sequence_number: seq += 1,
                       output_index:    output_index,
                       item:            function_call
                     })
  end

  out << sse_event('response.completed', {
                     type:            'response.completed',
                     sequence_number: seq + 1,
                     response:        {
                       id:         request_id,
                       object:     'response',
                       created_at: created_at,
                       status:     'completed',
                       model:      resolved_model,
                       output:     [completed_item, *function_calls],
                       usage:      usage
                     }
                   })

  log.info("[llm][api][openai][responses] action=stream_complete request_id=#{request_id} model=#{resolved_model}")
end

.token_aliases(key) ⇒ Object 



424
425
426
427
428
429
430
431
432
433
 
# File 'lib/legion/llm/api/openai/responses.rb', line 424

def self.token_aliases(key)
  case key.to_sym
  when :input, :input_tokens
    %i[input_tokens input]
  when :output, :output_tokens
    %i[output_tokens output]
  else
    [key.to_sym]
  end
end

.token_method(key) ⇒ Object 



435
436
437
438
439
440
441
442
 
# File 'lib/legion/llm/api/openai/responses.rb', line 435

def self.token_method(key)
  {
    input:         :input_tokens,
    input_tokens:  :input_tokens,
    output:        :output_tokens,
    output_tokens: :output_tokens
  }[key.to_sym]
end