[Responses API] Function calling

[Wauplin]

Built on top of #1576.

Based on https://platform.openai.com/docs/api-reference/responses/create and https://platform.openai.com/docs/guides/function-calling?api-mode=responses#streaming

Works both with and without streaming.

Note: implementation starts to be a completely messy, especially in streaming mode. Complexity increases as we add new event types. I do think a refactoring would be beneficial e.g. with an internal state object that keeps track of the current state and "knows" what to emit and when (typically to emit the "done"/"completed" events each time a new output/content is generated). Food for thoughts for a future PR.

Non-stream

Run

pnpm run example function

Output

{
  created_at: 1751467285177,
  error: null,
  id: 'resp_0b2ab98168a9813e0f7373f940221da4ef3211f43c9faac8',
  instructions: null,
  max_output_tokens: null,
  metadata: null,
  model: 'meta-llama/Llama-3.3-70B-Instruct',
  object: 'response',
  output: [
    {
      type: 'function_call',
      id: 'fc_f40ac964165602e2fcb2f955777acff8c4b9359d49eaf79b',
      call_id: '9cd167c7f',
      name: 'get_current_weather',
      arguments: '{"location": "Boston, MA", "unit": "fahrenheit"}',
      status: 'completed'
    }
  ],
  status: 'completed',
  tool_choice: 'auto',
  tools: [
    {
      name: 'get_current_weather',
      parameters: [Object],
      strict: true,
      type: 'function',
      description: 'Get the current weather in a given location'
    }
  ],
  temperature: 1,
  top_p: 1,
  output_text: ''
}

Stream

Run:

pnpm run example function_streaming

Output:

{
  type: 'response.created',
  response: {
    created_at: 1751467334073,
    error: null,
    id: 'resp_8d86745178f2b9fc0da000156655956181c76a7701712a05',
    instructions: null,
    max_output_tokens: null,
    metadata: null,
    model: 'meta-llama/Llama-3.3-70B-Instruct',
    object: 'response',
    output: [],
    status: 'in_progress',
    tool_choice: 'auto',
    tools: [ [Object] ],
    temperature: 1,
    top_p: 1
  },
  sequence_number: 0
}
{
  type: 'response.in_progress',
  response: {
    created_at: 1751467334073,
    error: null,
    id: 'resp_8d86745178f2b9fc0da000156655956181c76a7701712a05',
    instructions: null,
    max_output_tokens: null,
    metadata: null,
    model: 'meta-llama/Llama-3.3-70B-Instruct',
    object: 'response',
    output: [],
    status: 'in_progress',
    tool_choice: 'auto',
    tools: [ [Object] ],
    temperature: 1,
    top_p: 1
  },
  sequence_number: 1
}
{
  type: 'response.output_item.added',
  output_index: 0,
  item: {
    type: 'function_call',
    id: 'fc_9bdc8945b9cb6c95c5c248db4203f0707ba9fd338dee2454',
    call_id: '83a9d4baf',
    name: 'get_weather',
    arguments: ''
  },
  sequence_number: 2
}
{
  type: 'response.function_call_arguments.delta',
  item_id: 'fc_9bdc8945b9cb6c95c5c248db4203f0707ba9fd338dee2454',
  output_index: 0,
  delta: '{"latitude": 48.8567, "longitude": 2.3508}',
  sequence_number: 3
}
{
  type: 'response.function_call_arguments.done',
  item_id: 'fc_9bdc8945b9cb6c95c5c248db4203f0707ba9fd338dee2454',
  output_index: 0,
  arguments: '{"latitude": 48.8567, "longitude": 2.3508}',
  sequence_number: 4
}
{
  type: 'response.output_item.done',
  output_index: 0,
  item: {
    type: 'function_call',
    id: 'fc_9bdc8945b9cb6c95c5c248db4203f0707ba9fd338dee2454',
    call_id: '83a9d4baf',
    name: 'get_weather',
    arguments: '{"latitude": 48.8567, "longitude": 2.3508}',
    status: 'completed'
  },
  sequence_number: 5
}
{
  type: 'response.completed',
  response: {
    created_at: 1751467334073,
    error: null,
    id: 'resp_8d86745178f2b9fc0da000156655956181c76a7701712a05',
    instructions: null,
    max_output_tokens: null,
    metadata: null,
    model: 'meta-llama/Llama-3.3-70B-Instruct',
    object: 'response',
    output: [ [Object] ],
    status: 'completed',
    tool_choice: 'auto',
    tools: [ [Object] ],
    temperature: 1,
    top_p: 1
  },
  sequence_number: 6
}