Skip to content

OpenAI Provider

openai

codex_ai.providers.openai

Modern OpenAI adapter backed by the Responses API.

Requires: pip install codex-ai[openai]

Classes

OpenAIProvider

Async OpenAI Responses API adapter.

The provider exposes direct text, structured JSON, and streaming methods. Requests are not stored by default. The default model keeps the previous provider's cost-sensitive role while moving to the current GPT-5.6 family.

Parameters:

Name Type Description Default
api_key str | None

OpenAI API key. When omitted, the SDK reads OPENAI_API_KEY.

 None
model str

Default model used for requests.

 _DEFAULT_MODEL
client _OpenAIClientPort | None

Optional injected client implementing the Responses API port.

 None
store bool

Whether OpenAI may store responses. Defaults to False.

 False
reasoning_effort str | None

Default GPT-5.6 reasoning effort. Defaults to "none".

 'none'
timeout float

SDK request timeout in seconds.

 _DEFAULT_TIMEOUT_SECONDS
max_retries int

SDK retry count for retryable failures.

 _DEFAULT_MAX_RETRIES
base_url str | None

Optional OpenAI-compatible API base URL.

 None
Source code in src/codex_ai/providers/openai.py 
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
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
217
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
class OpenAIProvider:
    """
    Async OpenAI Responses API adapter.

    The provider exposes direct text, structured JSON, and streaming methods.
    Requests are not stored by default. The default model keeps the previous
    provider's cost-sensitive role while moving to the current GPT-5.6 family.

    Args:
        api_key: OpenAI API key. When omitted, the SDK reads ``OPENAI_API_KEY``.
        model: Default model used for requests.
        client: Optional injected client implementing the Responses API port.
        store: Whether OpenAI may store responses. Defaults to ``False``.
        reasoning_effort: Default GPT-5.6 reasoning effort. Defaults to ``"none"``.
        timeout: SDK request timeout in seconds.
        max_retries: SDK retry count for retryable failures.
        base_url: Optional OpenAI-compatible API base URL.
    """

    def __init__(
        self,
        api_key: str | None = None,
        model: str = _DEFAULT_MODEL,
        *,
        client: _OpenAIClientPort | None = None,
        store: bool = False,
        reasoning_effort: str | None = "none",
        timeout: float = _DEFAULT_TIMEOUT_SECONDS,
        max_retries: int = _DEFAULT_MAX_RETRIES,
        base_url: str | None = None,
    ) -> None:
        self._client = client or self._build_client(
            api_key=api_key,
            timeout=timeout,
            max_retries=max_retries,
            base_url=base_url,
        )
        self._request_mapper = _OpenAIRequestMapper(
            _RequestDefaults(
                model=model,
                store=store,
                reasoning_effort=reasoning_effort,
            )
        )

    async def answer(self, prompt: PromptResult, **kwargs: Any) -> str:
        """Generate text for the router/dispatcher contract."""
        return await self.generate_text(prompt, **kwargs)

    async def generate_text(
        self,
        prompt: PromptResult | str,
        *,
        model: str | None = None,
        **kwargs: Any,
    ) -> str:
        """Generate plain text through the Responses API."""
        request = self._request_mapper.build(
            prompt,
            model=model,
            runtime_options=kwargs,
        )
        try:
            response = await self._client.responses.create(**request)
        except OpenAIError as exc:
            raise self._provider_error(exc) from exc
        return getattr(response, "output_text", "") or ""

    async def generate_json(
        self,
        prompt: PromptResult | str,
        *,
        schema: type[BaseModel] | None = None,
        model: str | None = None,
        **kwargs: Any,
    ) -> Any:
        """
        Generate structured JSON.

        A Pydantic schema uses the SDK's native structured-output parser.
        Without a schema, JSON mode is used and the returned text is decoded.
        """
        request = self._request_mapper.build(
            prompt,
            model=model,
            runtime_options=kwargs,
        )

        if schema is not None:
            request["text_format"] = schema
            try:
                response = await self._client.responses.parse(**request)
            except OpenAIError as exc:
                raise self._provider_error(exc) from exc

            parsed = getattr(response, "output_parsed", None)
            if parsed is None:
                raise LLMProviderError("OpenAI Responses API returned no parsed output")
            return parsed

        text_options = dict(request.get("text") or {})
        text_options["format"] = {"type": "json_object"}
        request["text"] = text_options

        try:
            response = await self._client.responses.create(**request)
        except OpenAIError as exc:
            raise self._provider_error(exc) from exc

        output_text = getattr(response, "output_text", "") or ""
        try:
            return json.loads(output_text)
        except json.JSONDecodeError as exc:
            raise LLMProviderError(f"OpenAI returned invalid JSON: {exc}") from exc

    async def stream_text(
        self,
        prompt: PromptResult | str,
        *,
        model: str | None = None,
        **kwargs: Any,
    ) -> AsyncIterator[str]:
        """Yield typed text deltas from a streaming Responses API request."""
        request = self._request_mapper.build(
            prompt,
            model=model,
            runtime_options=kwargs,
        )
        request["stream"] = True

        try:
            stream = await self._client.responses.create(**request)
            async for event in stream:
                if getattr(event, "type", None) == "response.output_text.delta":
                    delta = getattr(event, "delta", "")
                    if delta:
                        yield delta
        except OpenAIError as exc:
            raise self._provider_error(exc) from exc

    @staticmethod
    def _build_client(
        *,
        api_key: str | None,
        timeout: float,
        max_retries: int,
        base_url: str | None,
    ) -> _OpenAIClientPort:
        client_options: dict[str, Any] = {
            "timeout": timeout,
            "max_retries": max_retries,
        }
        if api_key is not None:
            client_options["api_key"] = api_key
        if base_url is not None:
            client_options["base_url"] = base_url
        return cast(_OpenAIClientPort, AsyncOpenAI(**client_options))

    @staticmethod
    def _provider_error(exc: OpenAIError) -> LLMProviderError:
        request_id = getattr(exc, "request_id", None)
        request_context = f" (request_id={request_id})" if request_id else ""
        return LLMProviderError(f"OpenAI Responses API error{request_context}: {exc}")
Functions
answer(prompt, **kwargs) async

Generate text for the router/dispatcher contract.

Source code in src/codex_ai/providers/openai.py 
148
149
150
async def answer(self, prompt: PromptResult, **kwargs: Any) -> str:
    """Generate text for the router/dispatcher contract."""
    return await self.generate_text(prompt, **kwargs)
generate_text(prompt, *, model=None, **kwargs) async

Generate plain text through the Responses API.

Source code in src/codex_ai/providers/openai.py 
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
async def generate_text(
    self,
    prompt: PromptResult | str,
    *,
    model: str | None = None,
    **kwargs: Any,
) -> str:
    """Generate plain text through the Responses API."""
    request = self._request_mapper.build(
        prompt,
        model=model,
        runtime_options=kwargs,
    )
    try:
        response = await self._client.responses.create(**request)
    except OpenAIError as exc:
        raise self._provider_error(exc) from exc
    return getattr(response, "output_text", "") or ""
generate_json(prompt, *, schema=None, model=None, **kwargs) async

Generate structured JSON.

A Pydantic schema uses the SDK's native structured-output parser. Without a schema, JSON mode is used and the returned text is decoded.

Source code in src/codex_ai/providers/openai.py 
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
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
async def generate_json(
    self,
    prompt: PromptResult | str,
    *,
    schema: type[BaseModel] | None = None,
    model: str | None = None,
    **kwargs: Any,
) -> Any:
    """
    Generate structured JSON.

    A Pydantic schema uses the SDK's native structured-output parser.
    Without a schema, JSON mode is used and the returned text is decoded.
    """
    request = self._request_mapper.build(
        prompt,
        model=model,
        runtime_options=kwargs,
    )

    if schema is not None:
        request["text_format"] = schema
        try:
            response = await self._client.responses.parse(**request)
        except OpenAIError as exc:
            raise self._provider_error(exc) from exc

        parsed = getattr(response, "output_parsed", None)
        if parsed is None:
            raise LLMProviderError("OpenAI Responses API returned no parsed output")
        return parsed

    text_options = dict(request.get("text") or {})
    text_options["format"] = {"type": "json_object"}
    request["text"] = text_options

    try:
        response = await self._client.responses.create(**request)
    except OpenAIError as exc:
        raise self._provider_error(exc) from exc

    output_text = getattr(response, "output_text", "") or ""
    try:
        return json.loads(output_text)
    except json.JSONDecodeError as exc:
        raise LLMProviderError(f"OpenAI returned invalid JSON: {exc}") from exc
stream_text(prompt, *, model=None, **kwargs) async

Yield typed text deltas from a streaming Responses API request.

Source code in src/codex_ai/providers/openai.py 
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
async def stream_text(
    self,
    prompt: PromptResult | str,
    *,
    model: str | None = None,
    **kwargs: Any,
) -> AsyncIterator[str]:
    """Yield typed text deltas from a streaming Responses API request."""
    request = self._request_mapper.build(
        prompt,
        model=model,
        runtime_options=kwargs,
    )
    request["stream"] = True

    try:
        stream = await self._client.responses.create(**request)
        async for event in stream:
            if getattr(event, "type", None) == "response.output_text.delta":
                delta = getattr(event, "delta", "")
                if delta:
                    yield delta
    except OpenAIError as exc:
        raise self._provider_error(exc) from exc