pyagentic-core 2.2.2__tar.gz → 2.3.0a2__tar.gz

{pyagentic_core-2.2.2 → pyagentic_core-2.3.0a2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pyagentic-core
-Version: 2.2.2
+Version: 2.3.0a2
 Summary: Build LLM Agents in a Pythonic way
 Author-email: Ryan Mikulec <rmikulec.dev@gmail.com>
 License: MIT
@@ -21,6 +21,7 @@ Requires-Dist: c3linearize>=0.1.0
 Requires-Dist: anthropic>=0.62.0
 Requires-Dist: google-generativeai>=0.8.0
 Requires-Dist: transitions>=0.9.3
+Requires-Dist: pillow>=12.1.0
 Dynamic: license-file
 
 # PyAgentic

{pyagentic_core-2.2.2 → pyagentic_core-2.3.0a2}/pyagentic/_base/_agent/_agent.py

@@ -16,6 +16,7 @@ from typing import (
 
 from transitions import Machine
 from pydantic import BaseModel, ValidationError
+from PIL.Image import Image
 
 from pyagentic.logging import get_logger
 from pyagentic._base._tool import _ToolDefinition, tool
@@ -431,15 +432,27 @@ class BaseAgent(metaclass=AgentMeta):
             logger.exception(e)
             result = f"Tool `{tool_call.name}` failed: {e}. Please kindly state to the user that is failed, provide state, and ask if they want to try again."  # noqa E501
 
-        stringified_result = (
-            result.model_dump_json(indent=2)
-            if issubclass(result.__class__, BaseModel)
-            else str(result)
-        )
-        # Add tool result to conversation history for LLM
-        self.state._messages.append(
-            self.provider.to_tool_call_result_message(result=stringified_result, id_=tool_call.id)
-        )
+        # Handle image results - create a message with the image attached
+        if isinstance(result, Image):
+            stringified_result = f"[Image returned by {tool_call.name}]"
+            # Add tool result message
+            self.state._messages.append(
+                self.provider.to_tool_call_result_message(result=stringified_result, id_=tool_call.id)
+            )
+            # Add the image as a user message so the LLM can see it
+            self.state._messages.append(
+                Message(role="user", content=f"Image from {tool_call.name}:", image=result)
+            )
+        else:
+            stringified_result = (
+                result.model_dump_json(indent=2)
+                if issubclass(result.__class__, BaseModel)
+                else str(result)
+            )
+            # Add tool result to conversation history for LLM
+            self.state._messages.append(
+                self.provider.to_tool_call_result_message(result=stringified_result, id_=tool_call.id)
+            )
 
         if self.phases:
             self.state._update_state_machine(phases=self.phases)

pyagentic_core-2.3.0a2/pyagentic/_utils/_image.py

@@ -0,0 +1,20 @@
+from PIL.Image import Image
+import base64
+import io
+
+
+def _encode_image(image: Image) -> str:
+    """
+    Convert a PIL Image to a base64-encoded data URL.
+
+    Args:
+        image: PIL Image object to encode
+
+    Returns:
+        Base64-encoded data URL string
+    """
+    buffer = io.BytesIO()
+    image.save(buffer, format="PNG")
+    image_bytes = buffer.getvalue()
+    base64_image = base64.b64encode(image_bytes).decode("utf-8")
+    return f"data:image/png;base64,{base64_image}"

{pyagentic_core-2.2.2 → pyagentic_core-2.3.0a2}/pyagentic/llm/_openai.py

@@ -11,11 +11,13 @@ from openai.types.responses import ParsedResponse as OpenAIParsedResponse
 
 from typing import List, Optional, Type
 from pydantic import BaseModel
+from PIL.Image import Image
 
 from pyagentic._base._agent._agent_state import _AgentState
 from pyagentic._base._tool import _ToolDefinition
 from pyagentic.llm._provider import LLMProvider
 from pyagentic.models.llm import ProviderInfo, LLMResponse, ToolCall, Message, UsageInfo
+from pyagentic._utils._image import _encode_image
 
 
 class OpenAIMessage(Message):
@@ -85,6 +87,41 @@ class OpenAIProvider(LLMProvider):
         """
         return OpenAIMessage(type="function_call_output", call_id=id_, output=result)
 
+    def _convert_messages_to_openai_format(self, messages: List[Message]) -> List[dict]:
+        """
+        Convert pyagentic messages to OpenAI's API format.
+
+        Handles messages with images by converting them to OpenAI's content array format.
+
+        Args:
+            messages: List of Message objects from the agent state
+
+        Returns:
+            List of message dictionaries in OpenAI API format
+        """
+        openai_messages = []
+
+        for message in messages:
+            msg_dict = message.to_dict()
+
+            # If message has an image, convert to OpenAI's content array format
+            if message.image is not None:
+                image_url = _encode_image(message.image)
+
+                # Build content array with text and image
+                content = []
+                if message.content:
+                    content.append({"type": "text", "text": message.content})
+                content.append({"type": "image_url", "image_url": {"url": image_url}})
+
+                msg_dict["content"] = content
+                # Remove the image field as it's now in content
+                msg_dict.pop("image", None)
+
+            openai_messages.append(msg_dict)
+
+        return openai_messages
+
     async def generate(
         self,
         state: _AgentState,
@@ -113,11 +150,14 @@ class OpenAIProvider(LLMProvider):
         if tool_defs is None:
             tool_defs = []
 
+        # Convert messages to OpenAI format (handles images in messages)
+        input_messages = self._convert_messages_to_openai_format(state._messages)
+
         if response_format:
             response: OpenAIParsedResponse[Type[BaseModel]] = await self.client.responses.parse(
                 model=self._model,
                 instructions=state.system_message,
-                input=[message.to_dict() for message in state._messages],
+                input=input_messages,
                 tools=[tool.to_openai_spec() for tool in tool_defs],
                 text_format=response_format,
                 **kwargs,
@@ -142,7 +182,7 @@ class OpenAIProvider(LLMProvider):
             response: OpenAIResponse = await self.client.responses.create(
                 model=self._model,
                 instructions=state.system_message,
-                input=[message.to_dict() for message in state._messages],
+                input=input_messages,
                 tools=[tool.to_openai_spec() for tool in tool_defs],
                 **kwargs,
             )

{pyagentic_core-2.2.2 → pyagentic_core-2.3.0a2}/pyagentic/llm/_provider.py

@@ -8,6 +8,7 @@ compatible with the pyagentic framework.
 from typing import Optional, Type
 from abc import ABC, abstractmethod
 from pydantic import BaseModel
+from PIL.Image import Image
 
 from pyagentic._base._tool import _ToolDefinition
 from pyagentic._base._agent._agent_state import _AgentState
@@ -87,6 +88,9 @@ class LLMProvider(ABC):
         """
         Generate a response from the language model.
 
+        Images can be included by adding them to messages in the agent state.
+        Each provider handles image conversion to their API format.
+
         Args:
             state: The agent state containing conversation history and system messages
             tool_defs: Optional list of tool definitions the model can use

{pyagentic_core-2.2.2 → pyagentic_core-2.3.0a2}/pyagentic/models/llm.py

@@ -1,5 +1,8 @@
-from typing import Any, List, Optional
-from pydantic import BaseModel
+from typing import Any, List, Optional, TYPE_CHECKING
+from pydantic import BaseModel, ConfigDict
+
+if TYPE_CHECKING:
+    from PIL.Image import Image
 
 
 class Message(BaseModel):
@@ -10,9 +13,12 @@ class Message(BaseModel):
     to a dictionary format for API communication.
     """
 
+    model_config = ConfigDict(arbitrary_types_allowed=True)
+
     type: Optional[str] = None
     role: Optional[str] = None
     content: Optional[str] = None
+    image: Optional["Image"] = None
 
     def to_dict(self, exclude_none: bool = True):
         """

{pyagentic_core-2.2.2 → pyagentic_core-2.3.0a2}/pyagentic/models/response.py

@@ -1,10 +1,12 @@
-from pydantic import BaseModel, Field, create_model
+from pydantic import BaseModel, Field, create_model, field_serializer
 from typing import Type, Self, Union, Any
+from PIL.Image import Image
 
 from pyagentic._base._tool import _ToolDefinition
 from pyagentic._base._agent._agent_state import _AgentState
 
 from pyagentic._utils._typing import TypeCategory, analyze_type
+from pyagentic._utils._image import _encode_image
 from pyagentic.models.llm import ProviderInfo
 
 
@@ -20,6 +22,13 @@ class ToolResponse(BaseModel):
     call_depth: int
     output: Any
 
+    @field_serializer("output")
+    def serialize_output(self, value: Any, _info) -> Any:
+        """Serialize Pillow images to base64-encoded data URLs."""
+        if isinstance(value, Image):
+            return _encode_image(value)
+        return value
+
     @classmethod
     def from_tool_def(cls, tool_def: _ToolDefinition) -> Type[Self]:
         """
@@ -77,6 +86,13 @@ class AgentResponse(BaseModel):
     final_output: Union[str, Type[BaseModel]]
     provider_info: ProviderInfo
 
+    @field_serializer("final_output")
+    def serialize_final_output(self, value: Any, _info) -> Any:
+        """Serialize Pillow images to base64-encoded data URLs."""
+        if isinstance(value, Image):
+            return _encode_image(value)
+        return value
+
     @classmethod
     def from_agent_class(
         cls,

{pyagentic_core-2.2.2 → pyagentic_core-2.3.0a2}/pyagentic_core.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pyagentic-core
-Version: 2.2.2
+Version: 2.3.0a2
 Summary: Build LLM Agents in a Pythonic way
 Author-email: Ryan Mikulec <rmikulec.dev@gmail.com>
 License: MIT
@@ -21,6 +21,7 @@ Requires-Dist: c3linearize>=0.1.0
 Requires-Dist: anthropic>=0.62.0
 Requires-Dist: google-generativeai>=0.8.0
 Requires-Dist: transitions>=0.9.3
+Requires-Dist: pillow>=12.1.0
 Dynamic: license-file
 
 # PyAgentic

{pyagentic_core-2.2.2 → pyagentic_core-2.3.0a2}/pyagentic_core.egg-info/SOURCES.txt

@@ -17,6 +17,7 @@ pyagentic/_base/_agent/__init__.py
 pyagentic/_base/_agent/_agent.py
 pyagentic/_base/_agent/_agent_linking.py
 pyagentic/_base/_agent/_agent_state.py
+pyagentic/_utils/_image.py
 pyagentic/_utils/_typing.py
 pyagentic/_utils/_warnings.py
 pyagentic/llm/__init__.py

{pyagentic_core-2.2.2 → pyagentic_core-2.3.0a2}/pyagentic_core.egg-info/requires.txt

@@ -9,3 +9,4 @@ c3linearize>=0.1.0
 anthropic>=0.62.0
 google-generativeai>=0.8.0
 transitions>=0.9.3
+pillow>=12.1.0

{pyagentic_core-2.2.2 → pyagentic_core-2.3.0a2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "pyagentic-core"
-version = "2.2.2"
+version = "2.3.0-a.2"
 description = "Build LLM Agents in a Pythonic way"
 readme = "README.md"
 requires-python = ">=3.13"
@@ -24,6 +24,7 @@ dependencies = [
     "anthropic>=0.62.0",
     "google-generativeai>=0.8.0",
     "transitions>=0.9.3",
+    "pillow>=12.1.0",
 ]
 
 [dependency-groups]