prompt-caller 0.1.1__py3-none-any.whl → 0.2.0__py3-none-any.whl

prompt_caller/prompt_caller.py

@@ -1,13 +1,18 @@
 import os
 import re
+import ast
 
 import requests
 import yaml
 from dotenv import load_dotenv
 from jinja2 import Template
+from langgraph.types import Command
 from langchain_core.tools import tool
 from langchain_core.messages import HumanMessage, SystemMessage, ToolMessage
+from langchain.agents import create_agent
+from langchain.agents.middleware import wrap_tool_call
 from langchain_openai import ChatOpenAI
+from langchain_google_genai import ChatGoogleGenerativeAI
 from PIL import Image
 from pydantic import BaseModel, Field, create_model
 
@@ -18,7 +23,6 @@ load_dotenv()
 
 
 class PromptCaller:
-
     def __init__(self, promptPath="prompts"):
         self.promptPath = promptPath
 
@@ -39,17 +43,44 @@ class PromptCaller:
         template = Template(body)
         return template.render(context)
 
+    import re
+
     def _parseJSXBody(self, body):
         elements = []
-        tag_pattern = r"<(system|user|assistant|image)>(.*?)</\1>"
+        # 1. Regex to find tags, attributes string, and content
+        tag_pattern = r"<(system|user|assistant|image)([^>]*)>(.*?)</\1>"
+
+        # 2. Regex to find key="value" pairs within the attributes string
+        attr_pattern = r'(\w+)\s*=\s*"(.*?)"'
 
         matches = re.findall(tag_pattern, body, re.DOTALL)
 
-        for tag, content in matches:
-            elements.append({"role": tag, "content": content.strip()})
+        for tag, attrs_string, content in matches:
+            # 3. Parse the attributes string (e.g., ' tag="image 1"') into a dict
+            attributes = {}
+            if attrs_string:
+                attr_matches = re.findall(attr_pattern, attrs_string)
+                for key, value in attr_matches:
+                    attributes[key] = value
+
+            element = {"role": tag, "content": content.strip()}
+
+            # 4. Add the attributes to our element dict if they exist
+            if attributes:
+                element["attributes"] = attributes
+
+            elements.append(element)
 
         return elements
 
+    def _createChat(self, configuration):
+        if configuration.get("model") is not None and configuration.get(
+            "model"
+        ).startswith("gemini"):
+            return ChatGoogleGenerativeAI(**configuration)
+        else:
+            return ChatOpenAI(**configuration)
+
     def getImageBase64(self, url: str) -> str:
         response = requests.get(url)
         response.raise_for_status()
@@ -87,16 +118,18 @@ class PromptCaller:
                 if base64_image.startswith("http"):
                     base64_image = self.getImageBase64(base64_image)
 
-                messages.append(
-                    HumanMessage(
-                        content=[
-                            {
-                                "type": "image_url",
-                                "image_url": {"url": base64_image},
-                            }
-                        ]
-                    )
-                )
+                content = [
+                    {
+                        "type": "image_url",
+                        "image_url": {"url": base64_image},
+                    }
+                ]
+
+                tag = message.get("attributes", {}).get("tag")
+                if tag:
+                    content.append({"type": "text", "text": f"({tag})"})
+
+                messages.append(HumanMessage(content=content))
 
         return configuration, messages
 
@@ -110,7 +143,6 @@ class PromptCaller:
         return create_model("DynamicModel", **fields)
 
     def call(self, promptName, context=None):
-
         configuration, messages = self.loadPrompt(promptName, context)
 
         output = None
@@ -119,7 +151,7 @@ class PromptCaller:
             output = configuration.get("output")
             configuration.pop("output")
 
-        chat = ChatOpenAI(**configuration)
+        chat = self._createChat(configuration)
 
         if output:
             dynamicModel = self.createPydanticModel(output)
@@ -129,78 +161,123 @@ class PromptCaller:
 
         return response
 
-    def agent(self, promptName, context=None, tools=None, allowed_steps=3):
+    def _create_pdf_middleware(self):
+        """Middleware to handle tool responses that contain pdf content."""
+
+        @wrap_tool_call
+        def handle_pdf_response(request, handler):
+            # Execute the actual tool
+            result = handler(request)
+
+            # Check if result content is pdf data
+            if hasattr(result, "content"):
+                content = result.content
+                # Try to parse if it's a string representation of a list
+                if isinstance(content, str) and content.startswith("["):
+                    try:
+                        content = ast.literal_eval(content)
+                    except (ValueError, SyntaxError):
+                        pass
+
+                if (
+                    isinstance(content, list)
+                    and content
+                    and isinstance(content[0], dict)
+                    and "input_file" in content[0]
+                    and "pdf" in content[0]["file_data"]
+                ):
+                    # Use Command to add both tool result and image to messages
+                    return Command(
+                        update={"messages": [result, HumanMessage(content=content)]}
+                    )
 
-        configuration, messages = self.loadPrompt(promptName, context)
+            return result  # Return normal result
+
+        return handle_pdf_response
+
+    def _create_image_middleware(self):
+        """Middleware to handle tool responses that contain image content."""
+
+        @wrap_tool_call
+        def handle_image_response(request, handler):
+            # Execute the actual tool
+            result = handler(request)
+
+            # Check if result content is image data (list with image_url dict)
+            if hasattr(result, "content"):
+                content = result.content
+                # Try to parse if it's a string representation of a list
+                if isinstance(content, str) and content.startswith("["):
+                    try:
+                        content = ast.literal_eval(content)
+                    except (ValueError, SyntaxError):
+                        pass
+
+                if (
+                    isinstance(content, list)
+                    and content
+                    and isinstance(content[0], dict)
+                    and "image_url" in content[0]
+                ):
+                    # Use Command to add both tool result and image to messages
+                    return Command(
+                        update={"messages": [result, HumanMessage(content=content)]}
+                    )
 
-        output = None
+            return result  # Return normal result
 
-        if "output" in configuration:
-            output = configuration.get("output")
-            configuration.pop("output")
+        return handle_image_response
 
-            for message in messages:
-                if isinstance(message, SystemMessage):
-                    message.content += "\nOnly use the tool DynamicModel when providing an output call."
-                    break
+    def agent(
+        self, promptName, context=None, tools=None, output=None, allowed_steps=10
+    ):
+        configuration, messages = self.loadPrompt(promptName, context)
+
+        # Handle structured output from config
+        dynamicOutput = None
+        if output is None and "output" in configuration:
+            dynamicOutput = configuration.pop("output")
 
-        chat = ChatOpenAI(**configuration)
+        chat = self._createChat(configuration)
 
-        # Register the tools
+        # Prepare tools
         if tools is None:
             tools = []
-
-        # Transform functions in tools
         tools = [tool(t) for t in tools]
 
-        tools_dict = {t.name.lower(): t for t in tools}
-
+        # Handle response format (structured output)
+        response_format = None
         if output:
-            dynamicModel = self.createPydanticModel(output)
-
-        tools.extend([dynamicModel])
-        tools_dict["dynamicmodel"] = dynamicModel
-
-        chat = chat.bind_tools(tools)
-
-        try:
-            # First LLM invocation
-            response = chat.invoke(messages)
-            messages.append(response)
-
-            steps = 0
-            while response.tool_calls and steps < allowed_steps:
-                for tool_call in response.tool_calls:
-                    tool_name = tool_call["name"].lower()
-
-                    # If it's the final formatting tool, validate and return
-                    if tool_name == "dynamicmodel":
-                        return dynamicModel.model_validate(tool_call["args"])
-
-                    selected_tool = tools_dict.get(tool_name)
-                    if not selected_tool:
-                        raise ValueError(f"Unknown tool: {tool_name}")
-
-                    # Invoke the selected tool with provided arguments
-                    tool_response = selected_tool.invoke(tool_call)
-                    messages.append(tool_response)
-
-                # If the latest message is a ToolMessage, re-invoke the LLM
-                if isinstance(messages[-1], ToolMessage):
-                    response = chat.invoke(messages)
-                    messages.append(response)
-                else:
-                    break
-
-                steps += 1
-
-            # Final LLM call if the last message is still a ToolMessage
-            if isinstance(messages[-1], ToolMessage):
-                response = chat.invoke(messages)
-                messages.append(response)
+            response_format = output
+        elif dynamicOutput:
+            response_format = self.createPydanticModel(dynamicOutput)
+
+        # Extract system message for create_agent
+        system_prompt = None
+        user_messages = []
+        for msg in messages:
+            if isinstance(msg, SystemMessage):
+                system_prompt = msg.content
+            else:
+                user_messages.append(msg)
+
+        # Create and invoke agent
+        agent_graph = create_agent(
+            model=chat,
+            tools=tools,
+            system_prompt=system_prompt,
+            response_format=response_format,
+            middleware=[
+                self._create_image_middleware(),
+                self._create_pdf_middleware(),
+            ],
+        )
 
-            return response
+        result = agent_graph.invoke(
+            {"messages": user_messages}, config={"recursion_limit": allowed_steps}
+        )
 
-        except Exception as e:
-            # Replace with appropriate logging in production
-            raise RuntimeError("Error during agent process") from e
+        # Return structured output or last message
+        if response_format and result.get("structured_response"):
+            return result["structured_response"]
+        return result["messages"][-1]

{prompt_caller-0.1.1.dist-info → prompt_caller-0.2.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: prompt_caller
-Version: 0.1.1
+Version: 0.2.0
 Summary: This package is responsible for calling prompts in a specific format. It uses LangChain and OpenAI API
 Home-page: https://github.com/ThiNepo/prompt-caller
 Author: Thiago Nepomuceno
@@ -11,11 +11,13 @@ Classifier: Operating System :: OS Independent
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: pyyaml>=6.0.2
-Requires-Dist: python-dotenv>=1.0.1
+Requires-Dist: python-dotenv>=1.2.1
 Requires-Dist: Jinja2>=3.1.4
-Requires-Dist: langchain-openai>=0.3.5
-Requires-Dist: openai>=1.63.0
-Requires-Dist: pillow>=11.0.0
+Requires-Dist: langchain-core>=1.2.7
+Requires-Dist: langchain-openai>=1.1.7
+Requires-Dist: langchain-google-genai>=4.2.0
+Requires-Dist: openai>=2.16.0
+Requires-Dist: pillow>=12.1.0
 
 # PromptCaller
 

prompt_caller-0.2.0.dist-info/RECORD

@@ -0,0 +1,8 @@
+prompt_caller/__init__.py,sha256=4EGdeAJ_Ig7A-b-e17-nYbiXjckT7uL3to5lchMsoW4,41
+prompt_caller/__main__.py,sha256=dJ0dYtVmnhZuoV79R6YiAIta1ZkUKb-TEX4VEuYbgk0,139
+prompt_caller/prompt_caller.py,sha256=b6AvhCRDfSpRHpg5qGVkTV1WRwSsmq5l0uy79Y-XYEs,9798
+prompt_caller-0.2.0.dist-info/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
+prompt_caller-0.2.0.dist-info/METADATA,sha256=ntbB3PEOrASgd4UjhzXMSztBUIr3pdASL2R42mPNQak,4993
+prompt_caller-0.2.0.dist-info/WHEEL,sha256=eOLhNAGa2EW3wWl_TU484h7q1UNgy0JXjjoqKoxAAQc,92
+prompt_caller-0.2.0.dist-info/top_level.txt,sha256=iihiDRq-0VrKB8IKjxf7Lrtv-fLMq4tvgM4fH3x0I94,14
+prompt_caller-0.2.0.dist-info/RECORD,,

prompt_caller-0.1.1.dist-info/RECORD

@@ -1,8 +0,0 @@
-prompt_caller/__init__.py,sha256=4EGdeAJ_Ig7A-b-e17-nYbiXjckT7uL3to5lchMsoW4,41
-prompt_caller/__main__.py,sha256=dJ0dYtVmnhZuoV79R6YiAIta1ZkUKb-TEX4VEuYbgk0,139
-prompt_caller/prompt_caller.py,sha256=fy-pLXmYD2j5fnAxgBvxCNBrkQvDPGX0nWyqnaWeqSo,6737
-prompt_caller-0.1.1.dist-info/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
-prompt_caller-0.1.1.dist-info/METADATA,sha256=j1xmg_Y_NAhh7CmlCgAcfSMuHChZ9C4DPcszLADg7dk,4909
-prompt_caller-0.1.1.dist-info/WHEEL,sha256=eOLhNAGa2EW3wWl_TU484h7q1UNgy0JXjjoqKoxAAQc,92
-prompt_caller-0.1.1.dist-info/top_level.txt,sha256=iihiDRq-0VrKB8IKjxf7Lrtv-fLMq4tvgM4fH3x0I94,14
-prompt_caller-0.1.1.dist-info/RECORD,,