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

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

Overview

OpenAI /v1/responses (Responses API) client translator.

Per Phase 5:

- parse_request(body, env) -> Canonical::Request (handles
  input/instructions/reasoning/tools shapes specific to /v1/responses)
- format_response(canonical_or_pipeline_response) -> Hash with output[]
  containing {thinking, function_call, message} items
- format_error(error, status_code:, type:)
- events_emitter(out, ...) -> Events emitter conforming to the
  StreamAssembler contract.

Defined Under Namespace

Classes: Events

Constant Summary collapse

Canonical = 
Legion::Extensions::Llm::Canonical

Instance Method Summary collapse

Methods included from SharedExtractors

#apply_canonical_params_to_inference, #args_as_json_string, #args_as_object, #extract_content_text, #extract_thinking_text, #legion_routing_explicit_from_env, #legion_routing_from_env, #normalize_canonical_params, #normalize_response_format, #normalize_stop_sequences, #text_content_type?, #token_value

Instance Method Details

#build_inference_request(canonical_request, request_id:, server_caller:, modality: nil) ⇒ Object 



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
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 86

def build_inference_request(canonical_request, request_id:, server_caller:, modality: nil)
  tool_defs = build_tool_definitions(canonical_request.tools)

  extra = {}
  tier = canonical_request.metadata[:tier]
  extra[:tier] = tier.to_sym if tier
  routing_explicit = canonical_request.metadata[:routing_explicit]
  extra[:routing_explicit] = routing_explicit if routing_explicit

  messages = inference_messages(canonical_request.messages)
  request_kwargs = {
    id:              request_id,
    messages:        messages,
    system:          canonical_request.system,
    routing:         canonical_request.routing,
    tools:           tool_defs,
    tool_choice:     canonical_request.tool_choice,
    caller:          server_caller,
    conversation_id: canonical_request.conversation_id,
    stream:          canonical_request.stream == true,
    modality:        modality,
    thinking:        thinking_to_inference(canonical_request.thinking),
    cache:           { strategy: :default, cacheable: true },
    extra:           extra,
    metadata:        canonical_request.metadata.except(:upstream_body)
  }

  apply_canonical_params_to_inference(request_kwargs, canonical_request.params)

  Legion::LLM::Inference::Request.build(**request_kwargs)
end

#build_tool_choice(raw) ⇒ Object

OpenAI Responses tool_choice shapes:

"auto" / "none" / "required"          → corresponding sym
{type: 'function', name: 'X'}         → {type: 'function', name: 'X'}
{type: 'function', function: {name:}} → flatten to {type: 'function', name: 'X'}



122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 122

def build_tool_choice(raw)
  return nil if raw.nil?

  case raw
  when Hash
    symbolized = raw.transform_keys(&:to_sym)
    type = symbolized[:type].to_s
    if type == 'function'
      fn = symbolized[:function].is_a?(Hash) ? symbolized[:function].transform_keys(&:to_sym) : nil
      name = symbolized[:name] || fn&.[](:name)
      return { type: :function, name: name.to_s } if name

      symbolized
    else
      %w[auto none required any].include?(type) ? type.to_sym : symbolized
    end
  when String, Symbol
    raw.to_sym
  end
end

#ensure_reasoning_summary(body) ⇒ Object

When the caller asks for reasoning (‘reasoning.effort` set) but didn’t pin a summary mode, default to ‘summary: ’auto’‘. OpenAI’s /v1/responses lane omits reasoning summary content unless the request opts in — without this, codex→openai cells return only the message item (no reasoning), and the e2e validator reports “reasoning never produced” (P5-final-cells.md B3).



75
76
77
78
79
80
81
82
83
84
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 75

def ensure_reasoning_summary(body)
  reasoning = body[:reasoning]
  return body unless reasoning.is_a?(Hash)

  normalized = reasoning.transform_keys { |k| k.respond_to?(:to_sym) ? k.to_sym : k }
  return body.merge(reasoning: normalized) if normalized.key?(:summary)
  return body unless normalized[:effort]

  body.merge(reasoning: normalized.merge(summary: 'auto'))
end

#events_emitter(out, request_id:, model:, conv_id: nil) ⇒ Object 



188
189
190
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 188

def events_emitter(out, request_id:, model:, conv_id: nil)
  Events.new(out: out, request_id: request_id, model: model.to_s, conv_id: conv_id)
end

#format_chunk(canonical_chunk) ⇒ Object

Format a single Canonical::Chunk to a /v1/responses SSE event hash.



193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 193

def format_chunk(canonical_chunk)
  return nil if canonical_chunk.nil?

  case canonical_chunk.type
  when :text_delta
    { type: 'response.output_text.delta', delta: extract_content_text(canonical_chunk.delta),
      output_index: canonical_chunk.block_index || 0 }
  when :thinking_delta
    { type: 'response.thinking.delta', delta: canonical_chunk.delta.to_s,
      output_index: canonical_chunk.block_index || 0 }
  when :tool_call_delta
    format_tool_call_delta_chunk(canonical_chunk)
  when :done
    { type: 'response.completed' }
  end
end

#format_error(error, status_code: 500, type: 'server_error') ⇒ Object 



184
185
186
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 184

def format_error(error, status_code: 500, type: 'server_error')
  [status_code, { error: { message: error.respond_to?(:message) ? error.message : error.to_s, type: type } }]
end

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



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
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 143

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

  actionable_tool_calls = build_output_tool_calls(pipeline_response)
  server_tool_items = build_output_server_tool_items(pipeline_response)
  reasoning = build_output_reasoning(pipeline_response)

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

  # Responses protocol: a turn is always status `completed`. Both
  # client-callable calls (actionable_tool_calls) and LegionIO-run
  # results ride in output[] — client-callable ones as function_call
  # items the client executes and continues via function_call_output.
  # requires_action / action_required are Assistants API concepts that
  # real Responses clients (Codex) reject.
  {
    id:         request_id,
    object:     'response',
    created_at: Time.now.to_i,
    model:      resolved_model,
    output:     output,
    usage:      build_usage(tokens),
    status:     'completed'
  }
end

#format_tool_call_delta_chunk(canonical_chunk) ⇒ Object

G24 — server-executed tool chunks surface as completed function_call output items so the client sees the call name AND result inline. Plain client-callable chunks remain plain function_call_arguments.delta events.



214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 214

def format_tool_call_delta_chunk(canonical_chunk)
  tc = canonical_chunk.tool_call
  args = tc.respond_to?(:arguments) ? tc.arguments : {}
  output_index = canonical_chunk.block_index || 0

  if server_tool_chunk?(tc)
    {
      type:         'response.output_item.done',
      output_index: output_index,
      item:         {
        type:      'function_call',
        id:        "fc_#{SecureRandom.hex(12)}",
        call_id:   tc.respond_to?(:id) ? tc.id : nil,
        name:      tc.respond_to?(:name) ? tc.name.to_s : '',
        arguments: args_as_json_string(args),
        status:    'completed'
      }
    }
  else
    { type:         'response.function_call_arguments.delta',
      output_index: output_index,
      delta:        args_as_json_string(args) }
  end
end

#g24_format ⇒ Object

G24 — declares which execution-proxy contract shape this translator surfaces. Consumed by the lex-llm conformance shared examples.



34
35
36
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 34

def g24_format
  :openai_responses
end

#parse_request(body, env = {}) ⇒ Object 



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
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 38

def parse_request(body, env = {})
  log.debug('[llm][client_translator][openai_responses] action=parse_request')
  body = symbolize(body)

  messages = build_messages(body[:input], body[:instructions])
  tools = build_tools(body[:tools])
  params = build_params(body)
  thinking = build_thinking(body[:reasoning])
  tool_choice = build_tool_choice(body[:tool_choice])
  upstream_body = ensure_reasoning_summary(body)

  Canonical::Request.build(
    id:              env['HTTP_X_CLIENT_REQUEST_ID'] || "resp_#{SecureRandom.hex(16)}",
    messages:        messages,
    tools:           tools,
    tool_choice:     tool_choice,
    params:          params,
    thinking:        thinking,
    stream:          body[:stream] == true,
    conversation_id: env['HTTP_X_LEGION_CONVERSATION_ID'] || env['HTTP_THREAD_ID'] || body[:conversation],
    routing:         legion_routing_from_env(env),
    metadata:        {
      client_model:     body[:model],
      tier:             env['HTTP_X_LEGION_TIER'],
      routing_explicit: legion_routing_explicit_from_env(env),
      external_refs:    external_refs(body, env),
      upstream_body:    upstream_body # preserved for native call_responses path until executor is canonical
    }.compact
  )
end

#server_tool_chunk?(tool_call) ⇒ Boolean

Returns:

  • (Boolean) 


239
240
241
242
243
244
245
 
# File 'lib/legion/llm/api/client_translators/openai_responses.rb', line 239

def server_tool_chunk?(tool_call)
  source = tool_call.respond_to?(:source) ? tool_call.source : nil
  return false if source.nil?

  type = source.is_a?(Hash) ? (source[:type] || source['type']) : source
  %i[special registry extension mcp].include?(type&.to_sym)
end