ccs-llmconnector 1.1.2__py3-none-any.whl → 1.1.4__py3-none-any.whl

llmconnector/openai_client.py

@@ -4,164 +4,256 @@ from __future__ import annotations
 
 import base64
 import mimetypes
-from pathlib import Path
-import logging
-from typing import Optional, Sequence
-
-from openai import OpenAI
-
-from .types import ImageInput, MessageSequence, normalize_messages
-from .utils import clamp_retries, run_sync_in_thread, run_with_retries
-
-logger = logging.getLogger(__name__)
-
-
-class OpenAIResponsesClient:
+from pathlib import Path
+import logging
+from typing import Optional, Sequence
+
+from openai import OpenAI
+
+from .types import ImageInput, LLMResponse, MessageSequence, TokenUsage, normalize_messages
+from .utils import clamp_retries, run_sync_in_thread, run_with_retries
+
+logger = logging.getLogger(__name__)
+
+
+class OpenAIResponsesClient:
     """Convenience wrapper around the OpenAI Responses API."""
 
-    def generate_response(
-        self,
-        *,
-        api_key: str,
-        prompt: Optional[str] = None,
-        model: str,
-        max_tokens: int = 32000,
-        reasoning_effort: Optional[str] = None,
-        images: Optional[Sequence[ImageInput]] = None,
-        messages: Optional[MessageSequence] = None,
-        request_id: Optional[str] = None,
-        timeout_s: Optional[float] = None,
-        max_retries: Optional[int] = None,
-        retry_backoff_s: float = 0.5,
-    ) -> str:
-        """Generate a response from the specified model.
-
-        Args:
-            api_key: Secret key used to authenticate with OpenAI.
-            prompt: Natural-language instruction or query for the model.
-            model: Identifier of the OpenAI model to target (for example, ``"gpt-4o"``).
-            max_tokens: Cap for tokens across the entire exchange, defaults to 32000.
-            reasoning_effort: Optional reasoning effort hint (``"low"``, ``"medium"``, or ``"high"``).
-            images: Optional collection of image references (local paths or URLs).
-            messages: Optional list of chat-style messages (role/content).
-            request_id: Optional request identifier for tracing/logging.
-            timeout_s: Optional request timeout in seconds.
-            max_retries: Optional retry count for transient failures.
-            retry_backoff_s: Base delay (seconds) for exponential backoff between retries.
-
-        Returns:
-            The text output produced by the model.
-
+    def generate_response(
+        self,
+        *,
+        api_key: str,
+        prompt: Optional[str] = None,
+        model: str,
+        max_tokens: int = 32000,
+        reasoning_effort: Optional[str] = None,
+        images: Optional[Sequence[ImageInput]] = None,
+        messages: Optional[MessageSequence] = None,
+        request_id: Optional[str] = None,
+        timeout_s: Optional[float] = None,
+        max_retries: Optional[int] = None,
+        retry_backoff_s: float = 0.5,
+    ) -> str:
+        """Generate a response from the specified model.
+
+        Args:
+            api_key: Secret key used to authenticate with OpenAI.
+            prompt: Natural-language instruction or query for the model.
+            model: Identifier of the OpenAI model to target (for example, ``"gpt-4o"``).
+            max_tokens: Cap for tokens across the entire exchange, defaults to 32000.
+            reasoning_effort: Optional reasoning effort hint (``"low"``, ``"medium"``, or ``"high"``).
+            images: Optional collection of image references (local paths or URLs).
+            messages: Optional list of chat-style messages (role/content).
+            request_id: Optional request identifier for tracing/logging.
+            timeout_s: Optional request timeout in seconds.
+            max_retries: Optional retry count for transient failures.
+            retry_backoff_s: Base delay (seconds) for exponential backoff between retries.
+
+        Returns:
+            The text output produced by the model.
+
         Raises:
             ValueError: If required arguments are missing or the request payload is empty.
             OpenAIError: If the underlying OpenAI request fails.
-        """
-        if not api_key:
-            raise ValueError("api_key must be provided.")
-        if not prompt and not messages and not images:
-            raise ValueError("At least one of prompt, messages, or images must be provided.")
-        if not model:
-            raise ValueError("model must be provided.")
-
-        normalized_messages = normalize_messages(prompt=prompt, messages=messages)
-        content_messages: list[dict] = []
-        for message in normalized_messages:
-            content_blocks = []
-            if message["content"]:
-                content_blocks.append({"type": "input_text", "text": message["content"]})
-            content_messages.append({"role": message["role"], "content": content_blocks})
-
-        if images:
-            image_blocks = [self._to_image_block(image) for image in images]
-            target_index = next(
-                (
-                    index
-                    for index in range(len(content_messages) - 1, -1, -1)
-                    if content_messages[index]["role"] == "user"
-                ),
-                None,
-            )
-            if target_index is None:
-                content_messages.append({"role": "user", "content": image_blocks})
-            else:
-                content_messages[target_index]["content"].extend(image_blocks)
-
-        if not any(message["content"] for message in content_messages):
-            raise ValueError("No content provided for response generation.")
-
-        request_payload = {
-            "model": model,
-            "input": content_messages,
-            "max_output_tokens": max_tokens,
-        }
-
-        if reasoning_effort:
-            request_payload["reasoning"] = {"effort": reasoning_effort}
-
-        retry_count = clamp_retries(max_retries)
-
-        def _run_request() -> str:
-            client_kwargs = {"api_key": api_key}
-            if timeout_s is not None:
-                client_kwargs["timeout"] = timeout_s
-            client = OpenAI(**client_kwargs)
-            try:
-                response = client.responses.create(**request_payload)
-            except Exception as exc:  # Log and re-raise to preserve default behavior
-                logger.exception(
-                    "OpenAI Responses API request failed: %s request_id=%s",
-                    exc,
-                    request_id,
-                )
-                raise
-
-            output_text = response.output_text
-            logger.info(
-                "OpenAI generate_response succeeded: model=%s images=%d text_len=%d request_id=%s",
-                model,
-                len(images or []),
-                len(output_text or ""),
-                request_id,
-            )
-            return output_text
-
-        return run_with_retries(
-            func=_run_request,
-            max_retries=retry_count,
-            retry_backoff_s=retry_backoff_s,
-            request_id=request_id,
-        )
-
-    async def async_generate_response(
-        self,
-        *,
-        api_key: str,
-        prompt: Optional[str] = None,
-        model: str,
-        max_tokens: int = 32000,
-        reasoning_effort: Optional[str] = None,
-        images: Optional[Sequence[ImageInput]] = None,
-        messages: Optional[MessageSequence] = None,
-        request_id: Optional[str] = None,
-        timeout_s: Optional[float] = None,
-        max_retries: Optional[int] = None,
-        retry_backoff_s: float = 0.5,
-    ) -> str:
-        return await run_sync_in_thread(
-            lambda: self.generate_response(
-                api_key=api_key,
-                prompt=prompt,
-                model=model,
-                max_tokens=max_tokens,
-                reasoning_effort=reasoning_effort,
-                images=images,
-                messages=messages,
-                request_id=request_id,
-                timeout_s=timeout_s,
-                max_retries=max_retries,
-                retry_backoff_s=retry_backoff_s,
-            )
-        )
+        """
+        if not api_key:
+            raise ValueError("api_key must be provided.")
+        if not prompt and not messages and not images:
+            raise ValueError("At least one of prompt, messages, or images must be provided.")
+        if not model:
+            raise ValueError("model must be provided.")
+
+        normalized_messages = normalize_messages(prompt=prompt, messages=messages)
+        content_messages: list[dict] = []
+        for message in normalized_messages:
+            content_blocks = []
+            if message["content"]:
+                content_blocks.append({"type": "input_text", "text": message["content"]})
+            content_messages.append({"role": message["role"], "content": content_blocks})
+
+        if images:
+            image_blocks = [self._to_image_block(image) for image in images]
+            target_index = next(
+                (
+                    index
+                    for index in range(len(content_messages) - 1, -1, -1)
+                    if content_messages[index]["role"] == "user"
+                ),
+                None,
+            )
+            if target_index is None:
+                content_messages.append({"role": "user", "content": image_blocks})
+            else:
+                content_messages[target_index]["content"].extend(image_blocks)
+
+        if not any(message["content"] for message in content_messages):
+            raise ValueError("No content provided for response generation.")
+
+        request_payload = {
+            "model": model,
+            "input": content_messages,
+            "max_output_tokens": max_tokens,
+        }
+
+        if reasoning_effort:
+            request_payload["reasoning"] = {"effort": reasoning_effort}
+
+        retry_count = clamp_retries(max_retries)
+
+        def _run_request() -> str:
+            client_kwargs = {"api_key": api_key}
+            if timeout_s is not None:
+                client_kwargs["timeout"] = timeout_s
+            client = OpenAI(**client_kwargs)
+            try:
+                response = client.responses.create(**request_payload)
+            except Exception as exc:  # Log and re-raise to preserve default behavior
+                logger.exception(
+                    "OpenAI Responses API request failed: %s request_id=%s",
+                    exc,
+                    request_id,
+                )
+                raise
+
+            output_text = response.output_text
+            logger.info(
+                "OpenAI generate_response succeeded: model=%s images=%d text_len=%d request_id=%s",
+                model,
+                len(images or []),
+                len(output_text or ""),
+                request_id,
+            )
+            return output_text
+
+        return run_with_retries(
+            func=_run_request,
+            max_retries=retry_count,
+            retry_backoff_s=retry_backoff_s,
+            request_id=request_id,
+        )
+
+    def generate_response_with_usage(
+        self,
+        *,
+        api_key: str,
+        prompt: Optional[str] = None,
+        model: str,
+        max_tokens: int = 32000,
+        reasoning_effort: Optional[str] = None,
+        images: Optional[Sequence[ImageInput]] = None,
+        messages: Optional[MessageSequence] = None,
+        request_id: Optional[str] = None,
+        timeout_s: Optional[float] = None,
+        max_retries: Optional[int] = None,
+        retry_backoff_s: float = 0.5,
+    ) -> LLMResponse:
+        if not api_key:
+            raise ValueError("api_key must be provided.")
+        if not prompt and not messages and not images:
+            raise ValueError("At least one of prompt, messages, or images must be provided.")
+        if not model:
+            raise ValueError("model must be provided.")
+
+        normalized_messages = normalize_messages(prompt=prompt, messages=messages)
+        content_messages: list[dict] = []
+        for message in normalized_messages:
+            content_blocks = []
+            if message["content"]:
+                content_blocks.append({"type": "input_text", "text": message["content"]})
+            content_messages.append({"role": message["role"], "content": content_blocks})
+
+        if images:
+            image_blocks = [self._to_image_block(image) for image in images]
+            target_index = next(
+                (
+                    index
+                    for index in range(len(content_messages) - 1, -1, -1)
+                    if content_messages[index]["role"] == "user"
+                ),
+                None,
+            )
+            if target_index is None:
+                content_messages.append({"role": "user", "content": image_blocks})
+            else:
+                content_messages[target_index]["content"].extend(image_blocks)
+
+        if not any(message["content"] for message in content_messages):
+            raise ValueError("No content provided for response generation.")
+
+        request_payload = {
+            "model": model,
+            "input": content_messages,
+            "max_output_tokens": max_tokens,
+        }
+
+        if reasoning_effort:
+            request_payload["reasoning"] = {"effort": reasoning_effort}
+
+        retry_count = clamp_retries(max_retries)
+
+        def _run_request() -> LLMResponse:
+            client_kwargs = {"api_key": api_key}
+            if timeout_s is not None:
+                client_kwargs["timeout"] = timeout_s
+            client = OpenAI(**client_kwargs)
+            try:
+                response = client.responses.create(**request_payload)
+            except Exception as exc:  # Log and re-raise to preserve default behavior
+                logger.exception(
+                    "OpenAI Responses API request failed: %s request_id=%s",
+                    exc,
+                    request_id,
+                )
+                raise
+
+            output_text = response.output_text
+            usage = _extract_openai_usage(response)
+            logger.info(
+                "OpenAI generate_response_with_usage succeeded: model=%s images=%d text_len=%d request_id=%s",
+                model,
+                len(images or []),
+                len(output_text or ""),
+                request_id,
+            )
+            return LLMResponse(text=output_text, usage=usage, provider="openai", model=model)
+
+        return run_with_retries(
+            func=_run_request,
+            max_retries=retry_count,
+            retry_backoff_s=retry_backoff_s,
+            request_id=request_id,
+        )
+
+    async def async_generate_response(
+        self,
+        *,
+        api_key: str,
+        prompt: Optional[str] = None,
+        model: str,
+        max_tokens: int = 32000,
+        reasoning_effort: Optional[str] = None,
+        images: Optional[Sequence[ImageInput]] = None,
+        messages: Optional[MessageSequence] = None,
+        request_id: Optional[str] = None,
+        timeout_s: Optional[float] = None,
+        max_retries: Optional[int] = None,
+        retry_backoff_s: float = 0.5,
+    ) -> str:
+        return await run_sync_in_thread(
+            lambda: self.generate_response(
+                api_key=api_key,
+                prompt=prompt,
+                model=model,
+                max_tokens=max_tokens,
+                reasoning_effort=reasoning_effort,
+                images=images,
+                messages=messages,
+                request_id=request_id,
+                timeout_s=timeout_s,
+                max_retries=max_retries,
+                retry_backoff_s=retry_backoff_s,
+            )
+        )
 
     @staticmethod
     def _to_image_block(image: ImageInput) -> dict:
@@ -183,11 +275,11 @@ class OpenAIResponsesClient:
             "image_url": _encode_image_path(Path(image)),
         }
 
-    def generate_image(
-        self,
-        *,
-        api_key: str,
-        prompt: str,
+    def generate_image(
+        self,
+        *,
+        api_key: str,
+        prompt: str,
         model: str,
         image_size: Optional[str] = None,
         aspect_ratio: Optional[str] = None,
@@ -195,112 +287,164 @@ class OpenAIResponsesClient:
     ) -> bytes:
         """Generate an image using the OpenAI API.
 
-        Raises:
-            NotImplementedError: This method is not yet implemented for OpenAI.
-        """
-        raise NotImplementedError("Image generation is not implemented for OpenAI.")
-
-    async def async_generate_image(
-        self,
-        *,
-        api_key: str,
-        prompt: str,
-        model: str,
-        image_size: Optional[str] = None,
-        aspect_ratio: Optional[str] = None,
-        image: Optional[ImageInput] = None,
-    ) -> bytes:
-        return await run_sync_in_thread(
-            lambda: self.generate_image(
-                api_key=api_key,
-                prompt=prompt,
-                model=model,
-                image_size=image_size,
-                aspect_ratio=aspect_ratio,
-                image=image,
-            )
-        )
-
-    def list_models(
-        self,
-        *,
-        api_key: str,
-        request_id: Optional[str] = None,
-        timeout_s: Optional[float] = None,
-        max_retries: Optional[int] = None,
-        retry_backoff_s: float = 0.5,
-    ) -> list[dict[str, Optional[str]]]:
-        """Return the models available to the authenticated OpenAI account."""
-        if not api_key:
-            raise ValueError("api_key must be provided.")
-
-        retry_count = clamp_retries(max_retries)
-
-        def _run_request() -> list[dict[str, Optional[str]]]:
-            client_kwargs = {"api_key": api_key}
-            if timeout_s is not None:
-                client_kwargs["timeout"] = timeout_s
-            client = OpenAI(**client_kwargs)
-            try:
-                response = client.models.list()
-            except Exception as exc:
-                logger.exception(
-                    "OpenAI list models failed: %s request_id=%s", exc, request_id
-                )
-                raise
-            data = getattr(response, "data", []) or []
-
-            models: list[dict[str, Optional[str]]] = []
-            for model in data:
-                model_id = getattr(model, "id", None)
-                if model_id is None and isinstance(model, dict):
-                    model_id = model.get("id")
-                if not model_id:
-                    continue
-
-                display_name = getattr(model, "display_name", None)
-                if display_name is None and isinstance(model, dict):
-                    display_name = model.get("display_name")
-
-                models.append({"id": model_id, "display_name": display_name})
-
-            logger.info(
-                "OpenAI list_models succeeded: count=%d request_id=%s",
-                len(models),
-                request_id,
-            )
-            return models
-
-        return run_with_retries(
-            func=_run_request,
-            max_retries=retry_count,
-            retry_backoff_s=retry_backoff_s,
-            request_id=request_id,
-        )
-
-    async def async_list_models(
-        self,
-        *,
-        api_key: str,
-        request_id: Optional[str] = None,
-        timeout_s: Optional[float] = None,
-        max_retries: Optional[int] = None,
-        retry_backoff_s: float = 0.5,
-    ) -> list[dict[str, Optional[str]]]:
-        return await run_sync_in_thread(
-            lambda: self.list_models(
-                api_key=api_key,
-                request_id=request_id,
-                timeout_s=timeout_s,
-                max_retries=max_retries,
-                retry_backoff_s=retry_backoff_s,
-            )
-        )
-
-
-def _encode_image_path(path: Path) -> str:
+        Raises:
+            NotImplementedError: This method is not yet implemented for OpenAI.
+        """
+        raise NotImplementedError("Image generation is not implemented for OpenAI.")
+
+    async def async_generate_image(
+        self,
+        *,
+        api_key: str,
+        prompt: str,
+        model: str,
+        image_size: Optional[str] = None,
+        aspect_ratio: Optional[str] = None,
+        image: Optional[ImageInput] = None,
+    ) -> bytes:
+        return await run_sync_in_thread(
+            lambda: self.generate_image(
+                api_key=api_key,
+                prompt=prompt,
+                model=model,
+                image_size=image_size,
+                aspect_ratio=aspect_ratio,
+                image=image,
+            )
+        )
+
+    async def async_generate_response_with_usage(
+        self,
+        *,
+        api_key: str,
+        prompt: Optional[str] = None,
+        model: str,
+        max_tokens: int = 32000,
+        reasoning_effort: Optional[str] = None,
+        images: Optional[Sequence[ImageInput]] = None,
+        messages: Optional[MessageSequence] = None,
+        request_id: Optional[str] = None,
+        timeout_s: Optional[float] = None,
+        max_retries: Optional[int] = None,
+        retry_backoff_s: float = 0.5,
+    ) -> LLMResponse:
+        return await run_sync_in_thread(
+            lambda: self.generate_response_with_usage(
+                api_key=api_key,
+                prompt=prompt,
+                model=model,
+                max_tokens=max_tokens,
+                reasoning_effort=reasoning_effort,
+                images=images,
+                messages=messages,
+                request_id=request_id,
+                timeout_s=timeout_s,
+                max_retries=max_retries,
+                retry_backoff_s=retry_backoff_s,
+            )
+        )
+
+    def list_models(
+        self,
+        *,
+        api_key: str,
+        request_id: Optional[str] = None,
+        timeout_s: Optional[float] = None,
+        max_retries: Optional[int] = None,
+        retry_backoff_s: float = 0.5,
+    ) -> list[dict[str, Optional[str]]]:
+        """Return the models available to the authenticated OpenAI account."""
+        if not api_key:
+            raise ValueError("api_key must be provided.")
+
+        retry_count = clamp_retries(max_retries)
+
+        def _run_request() -> list[dict[str, Optional[str]]]:
+            client_kwargs = {"api_key": api_key}
+            if timeout_s is not None:
+                client_kwargs["timeout"] = timeout_s
+            client = OpenAI(**client_kwargs)
+            try:
+                response = client.models.list()
+            except Exception as exc:
+                logger.exception(
+                    "OpenAI list models failed: %s request_id=%s", exc, request_id
+                )
+                raise
+            data = getattr(response, "data", []) or []
+
+            models: list[dict[str, Optional[str]]] = []
+            for model in data:
+                model_id = getattr(model, "id", None)
+                if model_id is None and isinstance(model, dict):
+                    model_id = model.get("id")
+                if not model_id:
+                    continue
+
+                display_name = getattr(model, "display_name", None)
+                if display_name is None and isinstance(model, dict):
+                    display_name = model.get("display_name")
+
+                models.append({"id": model_id, "display_name": display_name})
+
+            logger.info(
+                "OpenAI list_models succeeded: count=%d request_id=%s",
+                len(models),
+                request_id,
+            )
+            return models
+
+        return run_with_retries(
+            func=_run_request,
+            max_retries=retry_count,
+            retry_backoff_s=retry_backoff_s,
+            request_id=request_id,
+        )
+
+    async def async_list_models(
+        self,
+        *,
+        api_key: str,
+        request_id: Optional[str] = None,
+        timeout_s: Optional[float] = None,
+        max_retries: Optional[int] = None,
+        retry_backoff_s: float = 0.5,
+    ) -> list[dict[str, Optional[str]]]:
+        return await run_sync_in_thread(
+            lambda: self.list_models(
+                api_key=api_key,
+                request_id=request_id,
+                timeout_s=timeout_s,
+                max_retries=max_retries,
+                retry_backoff_s=retry_backoff_s,
+            )
+        )
+
+
+def _encode_image_path(path: Path) -> str:
     """Generate a data URL for the provided image path."""
     data = path.read_bytes()
     encoded = base64.b64encode(data).decode("utf-8")
     mime_type = mimetypes.guess_type(path.name)[0] or "image/png"
-    return f"data:{mime_type};base64,{encoded}"
+    return f"data:{mime_type};base64,{encoded}"
+
+
+def _extract_openai_usage(response: object) -> TokenUsage | None:
+    usage_obj = getattr(response, "usage", None)
+    if usage_obj is None:
+        return None
+
+    input_tokens = getattr(usage_obj, "input_tokens", None)
+    output_tokens = getattr(usage_obj, "output_tokens", None)
+    total_tokens = getattr(usage_obj, "total_tokens", None)
+
+    if isinstance(usage_obj, dict):
+        input_tokens = usage_obj.get("input_tokens")
+        output_tokens = usage_obj.get("output_tokens")
+        total_tokens = usage_obj.get("total_tokens")
+
+    return TokenUsage(
+        input_tokens=int(input_tokens) if isinstance(input_tokens, int) else None,
+        output_tokens=int(output_tokens) if isinstance(output_tokens, int) else None,
+        total_tokens=int(total_tokens) if isinstance(total_tokens, int) else None,
+    )