Skip to content

Add Responses API #199

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
Konfekt wants to merge 1 commit into from
Closed

Add Responses API #199

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 15 additions & 0 deletions doc/vim-ai.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,6 +61,11 @@ Options: >
\ },
\}

To use the Responses API set `endpoint_url` to
`https://api.openai.com/v1/responses`. When using the Responses API,
`max_output_tokens` is supported and `max_completion_tokens`/`max_tokens`
are mapped if set.

Check OpenAI docs for more information:
https://platform.openai.com/docs/api-reference/completions

Expand Down Expand Up @@ -112,6 +117,11 @@ Options: >
\ },
\}

To use the Responses API set `endpoint_url` to
`https://api.openai.com/v1/responses`. When using the Responses API,
`max_output_tokens` is supported and `max_completion_tokens`/`max_tokens`
are mapped if set.

Check OpenAI docs for more information:
https://platform.openai.com/docs/api-reference/completions

Expand Down Expand Up @@ -165,6 +175,11 @@ Options: >
\ },
\}

To use the Responses API set `endpoint_url` to
`https://api.openai.com/v1/responses`. When using the Responses API,
`max_output_tokens` is supported and `max_completion_tokens`/`max_tokens`
are mapped if set.

Check OpenAI docs for more information:
https://platform.openai.com/docs/api-reference/chat

Expand Down
208 changes: 181 additions & 27 deletions py/providers/openai.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,8 @@ def _protocol_type_check(self) -> None:

def request(self, messages: Sequence[AIMessage]) -> Iterator[AIResponseChunk]:
options = self.options
openai_options = self._make_openai_options(options)
is_responses = self._is_responses_endpoint(options['endpoint_url'])
openai_options = self._make_responses_options(options) if is_responses else self._make_openai_options(options)
http_options = {
'request_timeout': options.get('request_timeout') or 20,
'auth_type': options['auth_type'],
Expand All @@ -41,42 +42,32 @@ def request(self, messages: Sequence[AIMessage]) -> Iterator[AIResponseChunk]:

def _flatten_content(messages):
# NOTE: Some providers like api.deepseek.com & api.groq.com expect a flat 'content' field.
flattened = []
for message in messages:
message = {**message}
if message['role'] in ('system', 'assistant'):
message['content'] = '\n'.join(map(lambda c: c['text'], message['content']))
return messages
flattened.append(message)
return flattened

url = options['endpoint_url']

if is_responses:
request = {
'input': self._make_responses_input(messages),
**openai_options
}
self.utils.print_debug("openai: [{}] request: {}", self.command_type, request)
response = self._openai_request(url, request, http_options)
return self._responses_chunks(response, openai_options)

request = {
'messages': _flatten_content(messages),
**openai_options
}
self.utils.print_debug("openai: [{}] request: {}", self.command_type, request)
url = options['endpoint_url']
response = self._openai_request(url, request, http_options)

_choice_key = 'delta' if openai_options.get('stream') else 'message'

def _get_delta(resp):
choices = resp.get('choices') or [{}]
return choices[0].get(_choice_key, {})

def _map_chunk(resp):
self.utils.print_debug("openai: [{}] response: {}", self.command_type, resp)
delta = _get_delta(resp)
if delta.get('reasoning_content'):
# NOTE: support for deepseek's reasoning_content
return {'type': 'thinking', 'content': delta.get('reasoning_content')}
if delta.get('reasoning'):
# NOTE: support for `reasoning` from openrouter
return {'type': 'thinking', 'content': delta.get('reasoning')}
if delta.get('content'):
return {'type': 'assistant', 'content': delta.get('content')}
return None # invalid chunk, this occured in deepseek models

def _filter_valid_chunks(chunk):
return chunk is not None

return filter(_filter_valid_chunks, map(_map_chunk, response))
return self._chat_completions_chunks(response, openai_options)

def _load_api_key(self):
raw_api_key = self.utils.load_api_key(
Expand Down Expand Up @@ -115,6 +106,7 @@ def _convert_option(name, converter):
_convert_option('stream', lambda x: bool(int(x)))
_convert_option('max_tokens', int)
_convert_option('max_completion_tokens', int)
_convert_option('max_output_tokens', int)
_convert_option('temperature', float)
_convert_option('frequency_penalty', float)
_convert_option('presence_penalty', float)
Expand Down Expand Up @@ -176,6 +168,168 @@ def _make_openai_options(self, options):

return result

def _make_responses_options(self, options):
result = {
'model': options['model'],
}

option_keys = [
'stream',
'temperature',
'top_p',
'seed',
'stop',
'reasoning',
]

for key in option_keys:
if key not in options:
continue

value = options[key]
if value == '':
continue

result[key] = value

max_output_tokens = options.get('max_output_tokens')
if max_output_tokens in ('', None, 0):
max_completion_tokens = options.get('max_completion_tokens')
if max_completion_tokens not in ('', None, 0):
max_output_tokens = max_completion_tokens
else:
max_tokens = options.get('max_tokens')
if max_tokens not in ('', None, 0):
max_output_tokens = max_tokens

if max_output_tokens not in ('', None, 0):
result['max_output_tokens'] = max_output_tokens

if 'reasoning' not in result:
reasoning_effort = options.get('reasoning_effort')
if reasoning_effort not in ('', None):
result['reasoning'] = { 'effort': reasoning_effort }

return result

def _make_responses_input(self, messages: Sequence[AIMessage]):
input_items = []
for message in messages:
role = message.get('role')
content = message.get('content') or []
if not content:
input_items.append({
'type': 'message',
'role': role,
'content': "",
})
continue
response_content = []
for part in content:
if part.get('type') == 'text':
response_content.append({
'type': 'input_text',
'text': part.get('text', ''),
})
elif part.get('type') == 'image_url':
image_url = part.get('image_url', {})
if image_url.get('url'):
response_content.append({
'type': 'input_image',
'image_url': image_url.get('url'),
})
input_items.append({
'type': 'message',
'role': role,
'content': response_content if response_content else "",
})
return input_items

def _responses_chunks(self, response: Iterator[Mapping[str, Any]], options: Mapping[str, Any]):
if options.get('stream'):
def _map_chunk(resp):
return self._map_responses_stream_event(resp)
return filter(lambda chunk: chunk is not None, map(_map_chunk, response))

def _non_stream_chunks():
resp = next(response, None)
if resp is None:
return
chunk = self._map_responses_response(resp)
if chunk is not None:
yield chunk
return _non_stream_chunks()

def _map_responses_stream_event(self, resp: Mapping[str, Any]):
self.utils.print_debug("openai: [{}] response: {}", self.command_type, resp)
event_type = resp.get('type')

if event_type == 'response.output_text.delta':
delta = resp.get('delta') or ''
return {'type': 'assistant', 'content': delta} if delta else None

if event_type == 'response.content_part.added':
part = resp.get('part') or {}
text = part.get('text') or ''
return {'type': 'assistant', 'content': text} if text else None

if event_type == 'error':
message = resp.get('message') or 'OpenAI Responses API error'
raise Exception(message)

return None

def _map_responses_response(self, resp: Mapping[str, Any]):
self.utils.print_debug("openai: [{}] response: {}", self.command_type, resp)
if resp.get('error'):
raise Exception(resp['error'])

output_text = resp.get('output_text')
if isinstance(output_text, str) and output_text:
return {'type': 'assistant', 'content': output_text}

output = resp.get('output') or []
text_parts = []
for item in output:
if item.get('type') == 'message':
for part in item.get('content', []):
if part.get('type') in ('output_text', 'text'):
text_parts.append(part.get('text', ''))
elif item.get('type') == 'output_text':
text_parts.append(item.get('text', ''))

if text_parts:
return {'type': 'assistant', 'content': ''.join(text_parts)}
return None

def _chat_completions_chunks(self, response, openai_options):
_choice_key = 'delta' if openai_options.get('stream') else 'message'

def _get_delta(resp):
choices = resp.get('choices') or [{}]
return choices[0].get(_choice_key, {})

def _map_chunk(resp):
self.utils.print_debug("openai: [{}] response: {}", self.command_type, resp)
delta = _get_delta(resp)
if delta.get('reasoning_content'):
# NOTE: support for deepseek's reasoning_content
return {'type': 'thinking', 'content': delta.get('reasoning_content')}
if delta.get('reasoning'):
# NOTE: support for `reasoning` from openrouter
return {'type': 'thinking', 'content': delta.get('reasoning')}
if delta.get('content'):
return {'type': 'assistant', 'content': delta.get('content')}
return None # invalid chunk, this occured in deepseek models

def _filter_valid_chunks(chunk):
return chunk is not None

return filter(_filter_valid_chunks, map(_map_chunk, response))

def _is_responses_endpoint(self, url: str) -> bool:
return '/responses' in url

def request_image(self, prompt: str) -> list[AIImageResponseChunk]:
options = self.options
http_options = {
Expand Down