mirascope 1.16.9__py3-none-any.whl → 1.17.0__py3-none-any.whl

mirascope/__init__.py

@@ -6,6 +6,15 @@ from contextlib import suppress
 with suppress(ImportError):
     from . import core as core
 
+from .core import (
+    BaseDynamicConfig,
+    BaseMessageParam,
+    BaseTool,
+    BaseToolKit,
+    Messages,
+    prompt_template,
+)
+
 with suppress(ImportError):
     from . import integrations as integrations
 
@@ -14,4 +23,14 @@ with suppress(ImportError):
 
 __version__ = importlib.metadata.version("mirascope")
 
-__all__ = ["core", "integrations", "retries", "__version__"]
+__all__ = [
+    "BaseDynamicConfig",
+    "BaseMessageParam",
+    "BaseTool",
+    "BaseToolKit",
+    "core",
+    "integrations",
+    "prompt_template",
+    "retries",
+    "__version__",
+]

mirascope/core/anthropic/_utils/_convert_message_params.py

@@ -5,6 +5,7 @@ import base64
 from anthropic.types import MessageParam
 
 from ...base import BaseMessageParam
+from ...base._utils._parse_content_template import _load_media, get_image_type
 
 
 def convert_message_params(
@@ -45,6 +46,18 @@ def convert_message_params(
                             },
                         }
                     )
+                elif part.type == "image_url":
+                    image = _load_media(part.url)
+                    converted_content.append(
+                        {
+                            "type": "image",
+                            "source": {
+                                "data": base64.b64encode(image).decode("utf-8"),
+                                "media_type": f"image/{get_image_type(image)}",
+                                "type": "base64",
+                            },
+                        }
+                    )
                 elif part.type == "document":
                     if part.media_type != "application/pdf":
                         raise ValueError(

mirascope/core/azure/_utils/_convert_message_params.py

@@ -51,6 +51,16 @@ def convert_message_params(
                             },
                         }
                     )
+                elif part.type == "image_url":
+                    converted_content.append(
+                        {
+                            "type": "image_url",
+                            "image_url": {
+                                "url": part.url,
+                                "detail": part.detail if part.detail else "auto",
+                            },
+                        }
+                    )
                 elif part.type == "tool_call":
                     converted_message_param = AssistantMessage(
                         tool_calls=[

mirascope/core/azure/_utils/_message_param_converter.py

@@ -4,18 +4,43 @@ from typing import cast
 from azure.ai.inference.models import (
     AssistantMessage,
     ChatRequestMessage,
+    ContentItem,
+    ImageContentItem,
     TextContentItem,
     ToolMessage,
     UserMessage,
 )
 
 from mirascope.core.azure._utils import convert_message_params
-from mirascope.core.base import BaseMessageParam, TextPart, ToolCallPart, ToolResultPart
+from mirascope.core.base import (
+    BaseMessageParam,
+    ImageURLPart,
+    TextPart,
+    ToolCallPart,
+    ToolResultPart,
+)
 from mirascope.core.base._utils._base_message_param_converter import (
     BaseMessageParamConverter,
 )
 
 
+def _parse_content(content: list[ContentItem]) -> list[TextPart | ImageURLPart]:
+    converted_parts = []
+    for part in content:
+        if isinstance(part, TextContentItem):
+            converted_parts.append(TextPart(type="text", text=part.text))
+        elif isinstance(part, ImageContentItem):
+            converted_parts.append(
+                ImageURLPart(
+                    type="image_url",
+                    url=part.image_url.url,
+                    detail=part.image_url.detail,
+                )
+            )
+
+    return converted_parts
+
+
 class AzureMessageParamConverter(BaseMessageParamConverter):
     """Converts between Azure `ChatRequestMessage` / `AssistantMessage` and Mirascope `BaseMessageParam`."""
 
@@ -41,19 +66,14 @@ class AzureMessageParamConverter(BaseMessageParamConverter):
                         BaseMessageParam(role="user", content=message_param.content)
                     )
                 elif isinstance(message_param.content, list):
-                    converted_parts = []
-                    for part in message_param.content:
-                        if isinstance(part, TextContentItem):
-                            converted_parts.append(
-                                TextPart(type="text", text=part.text)
-                            )
-                        # TODO: add support for image and audio parts here
+                    converted_parts = _parse_content(message_param.content)
                     converted.append(
                         BaseMessageParam(role="user", content=converted_parts)
                     )
             elif isinstance(message_param, AssistantMessage):
+                converted_parts = []
                 if tool_calls := message_param.tool_calls:
-                    content = [
+                    converted_parts.extend(
                         ToolCallPart(
                             type="tool_call",
                             name=tool_call.function.name,
@@ -61,10 +81,24 @@ class AzureMessageParamConverter(BaseMessageParamConverter):
                             args=json.loads(tool_call.function.arguments),
                         )
                         for tool_call in tool_calls
-                    ]
+                    )
+                if isinstance(message_param.content, str):
+                    converted_parts.append(
+                        TextPart(type="text", text=message_param.content)
+                    )
+                elif isinstance(message_param.content, list):
+                    converted_parts = _parse_content(message_param.content)
                 else:
-                    content = message_param.content or ""
-                converted.append(BaseMessageParam(role="assistant", content=content))
+                    converted_parts.append(TextPart(type="text", text=""))
+                converted.append(
+                    BaseMessageParam(
+                        role="assistant",
+                        content=converted_parts[0].text
+                        if len(converted_parts) == 1
+                        and isinstance(converted_parts[0], TextPart)
+                        else converted_parts,
+                    )
+                )
             elif isinstance(message_param, ToolMessage):
                 converted.append(
                     BaseMessageParam(

mirascope/core/base/__init__.py

@@ -12,10 +12,12 @@ from .from_call_args import FromCallArgs
 from .merge_decorators import merge_decorators
 from .message_param import (
     AudioPart,
+    AudioURLPart,
     BaseMessageParam,
     CacheControlPart,
     DocumentPart,
     ImagePart,
+    ImageURLPart,
     TextPart,
     ToolCallPart,
     ToolResultPart,
@@ -32,6 +34,7 @@ from .types import AudioSegment
 
 __all__ = [
     "AudioPart",
+    "AudioURLPart",
     "AudioSegment",
     "BaseCallKwargs",
     "BaseCallParams",
@@ -52,6 +55,7 @@ __all__ = [
     "FromCallArgs",
     "GenerateJsonSchemaNoTitles",
     "ImagePart",
+    "ImageURLPart",
     "merge_decorators",
     "metadata",
     "Messages",

mirascope/core/base/_utils/_convert_messages_to_message_params.py

@@ -10,10 +10,12 @@ from typing_extensions import TypeIs
 
 from ..message_param import (
     AudioPart,
+    AudioURLPart,
     BaseMessageParam,
     CacheControlPart,
     DocumentPart,
     ImagePart,
+    ImageURLPart,
     TextPart,
 )
 from ..types import AudioSegment, Image, has_pil_module, has_pydub_module
@@ -29,17 +31,33 @@ def _convert_message_sequence_part_to_content_part(
     | TextPart
     | CacheControlPart
     | ImagePart
+    | ImageURLPart
     | Image.Image
     | AudioPart
+    | AudioURLPart
     | AudioSegment
     | Wave_read
     | DocumentPart,
-) -> TextPart | ImagePart | AudioPart | CacheControlPart | DocumentPart:
+) -> (
+    TextPart
+    | ImagePart
+    | ImageURLPart
+    | AudioPart
+    | AudioURLPart
+    | CacheControlPart
+    | DocumentPart
+):
     if isinstance(message_sequence_part, str):
         return TextPart(text=message_sequence_part, type="text")
     elif isinstance(
         message_sequence_part,
-        TextPart | ImagePart | AudioPart | CacheControlPart | DocumentPart,
+        TextPart
+        | ImagePart
+        | ImageURLPart
+        | AudioPart
+        | AudioURLPart
+        | CacheControlPart
+        | DocumentPart,
     ):
         return message_sequence_part
     elif has_pil_module and isinstance(message_sequence_part, Image.Image):
@@ -82,13 +100,26 @@ def convert_message_content_to_message_param_content(
         | TextPart
         | CacheControlPart
         | ImagePart
+        | ImageURLPart
         | Image.Image
         | AudioPart
+        | AudioURLPart
         | AudioSegment
         | Wave_read
         | DocumentPart
     ],
-) -> list[TextPart | ImagePart | AudioPart | CacheControlPart | DocumentPart] | str:
+) -> (
+    list[
+        TextPart
+        | ImagePart
+        | ImageURLPart
+        | AudioPart
+        | AudioURLPart
+        | CacheControlPart
+        | DocumentPart
+    ]
+    | str
+):
     if isinstance(message_sequence, str):
         return message_sequence
     return [
@@ -113,8 +144,10 @@ def convert_messages_to_message_params(
         | TextPart
         | CacheControlPart
         | ImagePart
+        | ImageURLPart
         | Image.Image
         | AudioPart
+        | AudioURLPart
         | AudioSegment
         | Wave_read
         | DocumentPart

mirascope/core/base/_utils/_parse_content_template.py

@@ -9,10 +9,12 @@ from typing_extensions import TypedDict
 
 from ..message_param import (
     AudioPart,
+    AudioURLPart,
     BaseMessageParam,
     CacheControlPart,
     DocumentPart,
     ImagePart,
+    ImageURLPart,
     TextPart,
 )
 from ..types import Image, has_pil_module
@@ -138,7 +140,12 @@ def _load_media(source: str | bytes) -> bytes:
 
 def _construct_image_part(
     source: str | bytes | Image.Image, options: dict[str, str] | None
-) -> ImagePart:
+) -> ImagePart | ImageURLPart:
+    detail = None
+    if options:
+        detail = options.get("detail", None)
+    if isinstance(source, str) and source.startswith(("http://", "https://", "gs://")):
+        return ImageURLPart(type="image_url", url=source, detail=detail)
     if isinstance(source, Image.Image):
         image = pil_image_to_bytes(source)
         media_type = (
@@ -149,9 +156,6 @@ def _construct_image_part(
     else:
         image = _load_media(source)
         media_type = f"image/{get_image_type(image)}"
-    detail = None
-    if options:
-        detail = options.get("detail", None)
     return ImagePart(
         type="image",
         media_type=media_type,
@@ -160,8 +164,10 @@ def _construct_image_part(
     )
 
 
-def _construct_audio_part(source: str | bytes) -> AudioPart:
+def _construct_audio_part(source: str | bytes) -> AudioPart | AudioURLPart:
     # Note: audio does not currently support additional options, at least for now.
+    if isinstance(source, str) and source.startswith(("http://", "https://", "gs://")):
+        return AudioURLPart(type="audio_url", url=source)
     audio = _load_media(source)
     return AudioPart(
         type="audio", media_type=f"audio/{get_audio_type(audio)}", audio=audio
@@ -179,8 +185,16 @@ def _construct_document_part(source: str | bytes) -> DocumentPart:
 
 def _construct_parts(
     part: _Part, attrs: dict[str, Any]
-) -> list[TextPart | ImagePart | AudioPart | CacheControlPart | DocumentPart]:
-    if part["type"] == "image":
+) -> list[
+    TextPart
+    | ImagePart
+    | ImageURLPart
+    | AudioPart
+    | AudioURLPart
+    | CacheControlPart
+    | DocumentPart
+]:
+    if part["type"] in "image":
         source = attrs[part["template"]]
         return [_construct_image_part(source, part["options"])] if source else []
     elif part["type"] == "images":
@@ -229,7 +243,13 @@ def _construct_parts(
         source = attrs[part["template"]]
         if not isinstance(
             source,
-            TextPart | ImagePart | AudioPart | CacheControlPart | DocumentPart,
+            TextPart
+            | ImagePart
+            | ImageURLPart
+            | AudioPart
+            | AudioURLPart
+            | CacheControlPart
+            | DocumentPart,
         ):
             raise ValueError(
                 f"When using 'part' template, '{part['template']}' must be a valid content part."
@@ -246,7 +266,13 @@ def _construct_parts(
         for source in sources:
             if not isinstance(
                 source,
-                TextPart | ImagePart | AudioPart | CacheControlPart | DocumentPart,
+                TextPart
+                | ImagePart
+                | ImageURLPart
+                | AudioPart
+                | AudioURLPart
+                | CacheControlPart
+                | DocumentPart,
             ):
                 raise ValueError(
                     f"When using 'parts' template, '{part['template']}' must be a list of valid content parts."

mirascope/core/base/message_param.py

@@ -49,18 +49,44 @@ class ImagePart(BaseModel):
     detail: str | None
 
 
+class ImageURLPart(BaseModel):
+    """A content part for images with a URL or base64 encoded image data.
+
+    Attributes:
+        type: Always "image_url"
+        url: The URL to the image
+        detail: (Optional) The detail to use for the image (supported by OpenAI)
+    """
+
+    type: Literal["image_url"]
+    url: str
+    detail: str | None
+
+
 class AudioPart(BaseModel):
     """A content part for audio.
 
     Attributes:
         type: Always "audio"
         media_type: The media type (e.g. audio/wav)
-        audio: The raw audio bytes
+        audio: The raw audio bytes or base64 encoded audio data
     """
 
     type: Literal["audio"]
     media_type: str
-    audio: bytes
+    audio: bytes | str
+
+
+class AudioURLPart(BaseModel):
+    """A content part for audio with a URL or base64 encoded audio data.
+
+    Attributes:
+        type: Always "audio_url"
+        url: The URL to the audio
+    """
+
+    type: Literal["audio_url"]
+    url: str
 
 
 class DocumentPart(BaseModel):
@@ -125,7 +151,9 @@ class BaseMessageParam(BaseModel):
         | Sequence[
             TextPart
             | ImagePart
+            | ImageURLPart
             | AudioPart
+            | AudioURLPart
             | CacheControlPart
             | DocumentPart
             | ToolCallPart

mirascope/core/base/messages.py

@@ -10,10 +10,12 @@ from ._utils._convert_messages_to_message_params import (
 )
 from .message_param import (
     AudioPart,
+    AudioURLPart,
     BaseMessageParam,
     CacheControlPart,
     DocumentPart,
     ImagePart,
+    ImageURLPart,
     TextPart,
 )
 from .types import AudioSegment
@@ -27,8 +29,10 @@ class Messages:
             | TextPart
             | CacheControlPart
             | ImagePart
+            | ImageURLPart
             | Image.Image
             | AudioPart
+            | AudioURLPart
             | AudioSegment
             | Wave_read
             | DocumentPart
@@ -46,8 +50,10 @@ class Messages:
             | TextPart
             | CacheControlPart
             | ImagePart
+            | ImageURLPart
             | Image.Image
             | AudioPart
+            | AudioURLPart
             | AudioSegment
             | Wave_read
             | DocumentPart
@@ -67,8 +73,10 @@ class Messages:
             | TextPart
             | CacheControlPart
             | ImagePart
+            | ImageURLPart
             | Image.Image
             | AudioPart
+            | AudioURLPart
             | AudioSegment
             | Wave_read
             | DocumentPart
@@ -88,8 +96,10 @@ class Messages:
             | TextPart
             | CacheControlPart
             | ImagePart
+            | ImageURLPart
             | Image.Image
             | AudioPart
+            | AudioURLPart
             | AudioSegment
             | Wave_read
             | DocumentPart

mirascope/core/bedrock/_utils/_convert_message_params.py

@@ -3,6 +3,8 @@
 from typing import cast
 
 from ...base import BaseMessageParam
+from ...base._utils import get_image_type
+from ...base._utils._parse_content_template import _load_media
 from .._types import ConversationRoleType, InternalBedrockMessageParam
 
 
@@ -37,7 +39,22 @@ def convert_message_params(
                             " currently only supports JPEG, PNG, GIF, and WebP images."
                         )
                     converted_content.append(
-                        {"format": part.media_type, "bytes": part.image}
+                        {
+                            "image": {
+                                "format": part.media_type.split("/")[-1],
+                                "source": {"bytes": part.image},
+                            }
+                        }
+                    )
+                elif part.type == "image_url":
+                    image = _load_media(part.url)
+                    converted_content.append(
+                        {
+                            "image": {
+                                "format": get_image_type(image),
+                                "source": {"bytes": image},
+                            },
+                        }
                     )
                 elif part.type == "tool_result":
                     if converted_content:

mirascope/core/gemini/_utils/_convert_message_params.py

@@ -7,6 +7,8 @@ from google.generativeai import protos
 from google.generativeai.types import ContentDict
 
 from ...base import BaseMessageParam
+from ...base._utils import get_audio_type
+from ...base._utils._parse_content_template import _load_media
 
 
 def convert_message_params(
@@ -23,13 +25,9 @@ def convert_message_params(
                 )  # pragma: no cover
             converted_message_params += [
                 {
-                    "role": "user",
+                    "role": "system",
                     "parts": [message_param.content],
                 },
-                {
-                    "role": "model",
-                    "parts": ["Ok! I will adhere to this system message."],
-                },
             ]
         elif isinstance((content := message_param.content), str):
             converted_message_params.append(
@@ -55,6 +53,29 @@ def convert_message_params(
                         )
                     image = PIL.Image.open(io.BytesIO(part.image))
                     converted_content.append(image)
+                elif part.type == "image_url":
+                    if part.url.startswith(("https://", "http://")):
+                        image = PIL.Image.open(io.BytesIO(_load_media(part.url)))
+                        media_type = (
+                            PIL.Image.MIME[image.format]
+                            if image.format
+                            else "image/unknown"
+                        )
+                        if media_type not in [
+                            "image/jpeg",
+                            "image/png",
+                            "image/webp",
+                            "image/heic",
+                            "image/heif",
+                        ]:
+                            raise ValueError(
+                                f"Unsupported image media type: {media_type}. "
+                                "Gemini currently only supports JPEG, PNG, WebP, HEIC, "
+                                "and HEIF images."
+                            )
+                        converted_content.append(image)
+                    else:
+                        converted_content.append(protos.FileData(file_uri=part.url))
                 elif part.type == "audio":
                     if part.media_type not in [
                         "audio/wav",
@@ -72,6 +93,28 @@ def convert_message_params(
                     converted_content.append(
                         {"mime_type": part.media_type, "data": part.audio}
                     )
+                elif part.type == "audio_url":
+                    if part.url.startswith(("https://", "http://")):
+                        audio = _load_media(part.url)
+                        audio_type = get_audio_type(audio)
+                        if audio_type not in [
+                            "audio/wav",
+                            "audio/mp3",
+                            "audio/aiff",
+                            "audio/aac",
+                            "audio/ogg",
+                            "audio/flac",
+                        ]:
+                            raise ValueError(
+                                f"Unsupported audio media type: {audio_type}. "
+                                "Gemini currently only supports WAV, MP3, AIFF, AAC, OGG, "
+                                "and FLAC audio file types."
+                            )
+                        converted_content.append(
+                            {"mime_type": audio_type, "data": audio}
+                        )
+                    else:
+                        converted_content.append(protos.FileData(file_uri=part.url))
                 elif part.type == "tool_call":
                     converted_content.append(
                         protos.FunctionCall(