prompt-caller 0.1.4__py3-none-any.whl → 0.2.1__py3-none-any.whl

prompt_caller/prompt_caller.py

@@ -1,12 +1,16 @@
 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
@@ -157,86 +161,88 @@ class PromptCaller:
 
         return response
 
+    def _create_media_middleware(self):
+        """Middleware to handle tool responses that contain media content (images, PDFs)."""
+
+        @wrap_tool_call
+        def handle_media_response(request, handler):
+            result = handler(request)
+
+            if hasattr(result, "content"):
+                content = result.content
+
+                if isinstance(content, str) and content.startswith("["):
+                    try:
+                        content = ast.literal_eval(content)
+                    except (ValueError, SyntaxError):
+                        pass
+
+                # Check if content is media (image or PDF)
+                if (
+                    isinstance(content, list)
+                    and content
+                    and isinstance(content[0], dict)
+                ):
+                    is_media = (
+                        "image_url" in content[0]  # Image
+                        or ("input_file" == content[0].get("type", "").strip())  # PDF
+                    )
+                    if is_media:
+                        return Command(
+                            update={"messages": [result, HumanMessage(content=content)]}
+                        )
+
+            return result
+
+        return handle_media_response
+
     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.get("output")
-            configuration.pop("output")
-
-            for message in messages:
-                if isinstance(message, SystemMessage):
-                    message.content += "\n\nYou have to use the tool `dynamicmodel` when providing your final answer. If you don't, you have failed the task."
-                    break
+            dynamicOutput = configuration.pop("output")
 
         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:
-            tools.extend([output])
-            tools_dict[output.__name__.lower()] = output
+            response_format = output
         elif dynamicOutput:
-            dynamicModel = self.createPydanticModel(dynamicOutput)
-
-            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 dynamicOutput and tool_name == "dynamicmodel":
-                        return dynamicModel.model_validate(tool_call["args"])
-
-                    if output and tool_name == output.__name__.lower():
-                        return output.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 = 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_media_middleware()],
+        )
 
-            return response
+        result = agent_graph.invoke(
+            {"messages": user_messages}, config={"recursion_limit": allowed_steps}
+        )
 
-        except Exception as e:
-            print(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-0.1.4.dist-info → prompt_caller-0.2.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: prompt_caller
-Version: 0.1.4
+Version: 0.2.1
 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
@@ -11,12 +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: langchain-google-genai==2.1.5
-Requires-Dist: openai>=1.63.0
-Requires-Dist: pillow>=11.0.0
+Requires-Dist: langchain>=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-0.2.1.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=U2K_qMMuTBP94cjdac7gPhac8Y6DGOSClfIBFuwpztg,8356
+prompt_caller-0.2.1.dist-info/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
+prompt_caller-0.2.1.dist-info/METADATA,sha256=b7cSmJA138gThOs1rB298gyS_yFPmbA4OlDfAAEXqL0,4988
+prompt_caller-0.2.1.dist-info/WHEEL,sha256=eOLhNAGa2EW3wWl_TU484h7q1UNgy0JXjjoqKoxAAQc,92
+prompt_caller-0.2.1.dist-info/top_level.txt,sha256=iihiDRq-0VrKB8IKjxf7Lrtv-fLMq4tvgM4fH3x0I94,14
+prompt_caller-0.2.1.dist-info/RECORD,,

prompt_caller-0.1.4.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=0Rqfzp7hUgPH10auxEwJMoaASOSZNlTyDvS8Hd6X0Yk,8333
-prompt_caller-0.1.4.dist-info/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
-prompt_caller-0.1.4.dist-info/METADATA,sha256=0x0EKp50KvL-dYuc9n8TEvouL53jD_V1fMw2bSKGPEY,4955
-prompt_caller-0.1.4.dist-info/WHEEL,sha256=eOLhNAGa2EW3wWl_TU484h7q1UNgy0JXjjoqKoxAAQc,92
-prompt_caller-0.1.4.dist-info/top_level.txt,sha256=iihiDRq-0VrKB8IKjxf7Lrtv-fLMq4tvgM4fH3x0I94,14
-prompt_caller-0.1.4.dist-info/RECORD,,