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

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

Overview

Sinatra extension for /v1/responses — parse → translate → execute → respond. All translation lives in API::ClientTranslators::OpenAIResponses.

Class Method Summary collapse

Class Method Details

.flush_pending(messages, pending) ⇒ Object 



227
228
229
230
231
232
233
234
235
236
237
238
 
# File 'lib/legion/llm/api/namespaces/openai/responses.rb', line 227

def self.flush_pending(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

.normalize_input_array(input) ⇒ Object

Helper kept at module-level for the input_tokens/count handler (and as a public seam for tests). Mirrors the translator’s internal normalization but stays callable without instantiating a translator.



195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
 
# File 'lib/legion/llm/api/namespaces/openai/responses.rb', line 195

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

  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 << {
        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(messages, pending)
      messages << { role: 'tool', tool_call_id: item[:call_id], content: item[:output].to_s }
    else
      flush_pending(messages, pending)
      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(messages, pending)
  messages
end

.registered(app) ⇒ Object

rubocop:disable Metrics/AbcSize



20
21
22
23
24
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
120
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
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
 
# File 'lib/legion/llm/api/namespaces/openai/responses.rb', line 20

def self.registered(app) # rubocop:disable Metrics/AbcSize
  log.debug('[llm][api][namespaces][openai][responses] registering routes')

  app.post '/v1/responses' do
    require_llm!
    request_started_at = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC)
    body = parse_request_body

    input = body[:input]
    unless input.is_a?(Array) || input.is_a?(String)
      return openai_error('input is required (string or array)',
                          type: 'invalid_request_error', status_code: 400)
    end

    translator = Legion::LLM::API::ClientTranslators::OpenAIResponses.new
    canonical_request = translator.parse_request(body, env)
    # Default reasoning.summary to 'auto' when the caller asked
    # for reasoning but didn't pin a summary mode — OpenAI's
    # /v1/responses lane omits reasoning content otherwise (B3).
    body = translator.ensure_reasoning_summary(body)
    request_id = canonical_request.id
    model = body[:model] || Legion::Settings[:llm][:default_model] || 'default'
    streaming = canonical_request.stream

    inference_request = translator.build_inference_request(
      canonical_request,
      request_id:    request_id,
      server_caller: build_server_caller(source: 'openai_responses', path: request.path, env: env)
    )

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

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

    canonical_format = Legion::LLM::API::DebugFormats.canonical_format?(env)
    echo_request = Legion::LLM::API::DebugFormats.echo_request?(env)

    if streaming
      content_type 'text/event-stream'
      headers 'Cache-Control' => 'no-cache', 'Connection' => 'keep-alive', 'X-Accel-Buffering' => 'no'
      stream do |out|
        emitter = if canonical_format
                    Legion::LLM::API::DebugFormats.canonical_event_emitter(out)
                  else
                    translator.events_emitter(out, request_id: request_id, model: model)
                  end
        Legion::LLM::API::DebugFormats.emit_echo_request_sse(out, canonical_request) if echo_request

        assembler = Legion::LLM::API::StreamAssembler.new(
          emitter:    emitter,
          request_id: request_id,
          model:      model
        )
        pipeline_response = if executor.respond_to?(:call_responses)
                              executor.call_responses(body: body, stream: true) { |c| assembler.push(c) }
                            else
                              executor.call_stream { |c| assembler.push(c) }
                            end
        assembler.finalize(pipeline_response)
        log_api_completion_summary(
          namespace:         'namespaces][openai][responses',
          request_id:        request_id,
          pipeline_response: pipeline_response,
          stream:            true,
          started_at:        request_started_at
        )
      rescue Legion::LLM::API::StreamAssembler::StreamClosed
        # Client disconnected — caller treats as cancellation per G10.
      rescue IOError, Errno::EPIPE
        # Client disconnected mid-write before assembler caught it.
      rescue StandardError => e
        handle_exception(e, level: :error, handled: false,
                            operation: 'llm.api.namespaces.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 = if executor.respond_to?(:call_responses)
                            executor.call_responses(body: body, stream: false)
                          else
                            executor.call
                          end
      log_api_completion_summary(
        namespace:         'namespaces][openai][responses',
        request_id:        request_id,
        pipeline_response: pipeline_response,
        stream:            false,
        started_at:        request_started_at
      )

      if canonical_format
        status_code, response_headers, body_string = Legion::LLM::API::DebugFormats.render_canonical_response(
          pipeline_response, canonical_request: canonical_request, env: env
        )
        status status_code
        response_headers.each { |k, v| headers k => v }
        body_string
      else
        formatted = translator.format_response(pipeline_response, request_id: request_id, model: model)
        formatted = Legion::LLM::API::DebugFormats.attach_echo_request(formatted, canonical_request) if echo_request
        content_type :json
        status 200
        Legion::JSON.dump(formatted)
      end
    end
  rescue Legion::LLM::AuthError => e
    handle_exception(e, level: :error, handled: true, operation: 'llm.api.namespaces.openai.responses.auth')
    openai_error(e.message, type: 'authentication_error', status_code: 401)
  rescue Legion::LLM::RateLimitError => e
    handle_exception(e, level: :warn, handled: true, operation: 'llm.api.namespaces.openai.responses.rate_limit')
    openai_error(e.message, type: 'rate_limit_error', code: 'rate_limit_exceeded', status_code: 429)
  rescue Legion::LLM::ProviderDown, Legion::LLM::ProviderError => e
    handle_exception(e, level: :error, handled: true, operation: 'llm.api.namespaces.openai.responses.provider')
    openai_error(e.message, type: 'server_error', status_code: 502)
  rescue StandardError => e
    handle_exception(e, level: :error, handled: false, operation: 'llm.api.namespaces.openai.responses')
    openai_error(e.message, type: 'server_error', status_code: 500)
  end

  app.get '/v1/responses/:id' do
    openai_error("Response '#{params[:id]}' not found", type: 'invalid_request_error',
                                                        code: 'response_not_found', status_code: 404)
  end

  app.delete '/v1/responses/:id' do
    content_type :json
    Legion::JSON.dump({ id: params[:id], object: 'response', deleted: true })
  end

  app.post '/v1/responses/:id/cancel' do
    openai_error("Response '#{params[:id]}' not found or already completed",
                 type: 'invalid_request_error', status_code: 404)
  end

  app.get '/v1/responses/:id/input_items' do
    content_type :json
    Legion::JSON.dump({ object: 'list', data: [], has_more: false })
  end

  app.post '/v1/responses/:id/input_tokens/count' do
    body  = parse_request_body
    input = body[:input]
    model = body[:model] || params[:id]
    messages = case input
               when Array  then Responses.normalize_input_array(input)
               when String then [{ role: 'user', content: input }]
               else []
               end
    result = Legion::LLM::TokenEstimation.estimate(messages: messages, model: model.to_s)
    content_type :json
    Legion::JSON.dump(result)
  rescue StandardError => e
    handle_exception(e, level: :error, handled: false,
                        operation: 'llm.api.namespaces.openai.responses.count_tokens')
    openai_error(e.message, type: 'server_error', status_code: 500)
  end

  app.post '/v1/responses/:id/compact' do
    openai_error("Response '#{params[:id]}' not found", type: 'invalid_request_error', status_code: 404)
  end

  app.post '/api/llm/inference/v1/responses' do
    call env.merge('PATH_INFO' => '/v1/responses')
  end

  log.debug('[llm][api][namespaces][openai][responses] routes registered')
rescue StandardError => e
  handle_exception(e, level: :error, handled: false,
                      operation: 'llm.api.namespaces.openai.responses.register')
end