From d9de70555ada7a7a8c231cc7a1a145ce69ca0e4b Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=D0=93=D0=BB=D0=B5=D0=B1=20=D0=9D=D0=B8=D0=BA=D0=B8=D1=88?= =?UTF-8?q?=D0=B8=D0=BD?= Date: Thu, 11 Jun 2026 16:14:55 +0000 Subject: [PATCH] Add main.py --- main.py | 167 +++++++++++++------------------------------------------- 1 file changed, 37 insertions(+), 130 deletions(-) diff --git a/main.py b/main.py index b65101c..52726c8 100644 --- a/main.py +++ b/main.py @@ -1,15 +1,13 @@ import os import asyncio -from typing import TypedDict, Dict - -from langchain_openai import ChatOpenAI +from langchain_openai import ChatOpenAI, OpenAIEmbeddings +from langchain_chroma import Chroma from langchain_core.messages import HumanMessage from langchain.tools import tool -from langgraph.graph import StateGraph, START, END -from pydantic import BaseModel, Field -from langchain_core.output_parsers import PydanticOutputParser +from deepagents import create_deep_agent +from deepagents.backends import FilesystemBackend, LocalShellBackend, CompositeBackend -# LLM setup +# LLM configuration – always OpenRouter llm = ChatOpenAI( model="openai/gpt-oss-20b:free", base_url="https://openrouter.ai/api/v1", @@ -17,138 +15,47 @@ llm = ChatOpenAI( temperature=0.0, ) -# State definition -class CodeReviewState(TypedDict): - code: str - draft_review: str - criteria_scores: Dict[str, int] - weakest_criterion: str - verdict: str - round: int - max_rounds: int +# Embeddings and vector store for RAG +embeddings = OpenAIEmbeddings( + model="text-embedding-3-small", + base_url="https://openrouter.ai/api/v1", + api_key=os.getenv("OPENAI_API_KEY"), +) +vector_store = Chroma(collection_name="knowledge", embedding_function=embeddings) -# Reflect output model -class ReflectOutput(BaseModel): - pep8: int = Field(description="Score 0-10 for PEP8 compliance") - type_hints: int = Field(description="Score 0-10 for type hints usage") - edge_cases: int = Field(description="Score 0-10 for edge case coverage") - naming: int = Field(description="Score 0-10 for naming conventions") - weakest_criterion: str = Field(description="Criterion with lowest score") - verdict: str = Field(description="'ok' or 'needs_revision'") - -reflect_parser = PydanticOutputParser(pydantic_object=ReflectOutput) - -# Dummy tool for agent -@tool -def echo_tool(query: str) -> str: - return query - -# Agent creation -from deepagents import create_deep_agent -from deepagents.backends import FilesystemBackend, LocalShellBackend, CompositeBackend +# Backend for file operations – virtual mode so no real files are created backend = CompositeBackend( default=LocalShellBackend(root_dir="./workspace", virtual_mode=True, inherit_env=True), routes={}, ) + +# Tool that performs a vector search and returns the top 3 snippets +@tool +def rag_search(query: str) -> str: + """Search the vector store for relevant documents and return a concise summary.""" + docs = vector_store.similarity_search(query, k=3) + if not docs: + return "No relevant information found." + snippets = "\n---\n".join(doc.page_content for doc in docs) + return f"Top matches:\n{snippets}" + +# Create the deep agent with the RAG tool agent = create_deep_agent( model=llm, - tools=[echo_tool], + tools=[rag_search], backend=backend, - system_prompt="You are a code review assistant.", + system_prompt="You are a helpful assistant that uses a knowledge base to answer questions. Use the rag_search tool when you need external information.", ) -# Node functions -async def draft_review(state: CodeReviewState) -> CodeReviewState: - prompt = f"""Write a concise code review (3-6 points) for the following Python function. Focus on style, correctness, and potential improvements. - -```python -{state['code']} -``` - -Return only the review text.""" - response = await agent.ainvoke({"messages": [HumanMessage(content=prompt)]}, {"configurable": {"thread_id": "draft-review"}}) - review_text = response["messages"][-1].content - state["draft_review"] = review_text - return state - -async def reflect(state: CodeReviewState) -> CodeReviewState: - prompt = f"""You are a senior reviewer. Evaluate the following review text against four criteria: PEP8, type hints, edge cases, naming. Assign each a score 0-10. Identify the weakest criterion and give a verdict: 'ok' if all scores >=7, else 'needs_revision'. Return a JSON with keys: pep8, type_hints, edge_cases, naming, weakest_criterion, verdict. - -Review: -{state['draft_review']}""" - response = await agent.ainvoke({"messages": [HumanMessage(content=prompt)]}, {"configurable": {"thread_id": "reflect"}}) - json_text = response["messages"][-1].content - try: - parsed = reflect_parser.parse(json_text) - except Exception: - parsed = ReflectOutput(pep8=5, type_hints=5, edge_cases=5, naming=5, weakest_criterion="pep8", verdict="needs_revision") - state["criteria_scores"] = { - "pep8": parsed.pep8, - "type_hints": parsed.type_hints, - "edge_cases": parsed.edge_cases, - "naming": parsed.naming, - } - state["weakest_criterion"] = parsed.weakest_criterion - state["verdict"] = parsed.verdict - return state - -async def rewrite(state: CodeReviewState) -> CodeReviewState: - crit = state["weakest_criterion"] - prompt = f"""Improve the review section that addresses the weakest criterion '{crit}'. Provide a more detailed point for that criterion. Keep the rest of the review unchanged. - -Current review: -{state['draft_review']}""" - response = await agent.ainvoke({"messages": [HumanMessage(content=prompt)]}, {"configurable": {"thread_id": "rewrite"}}) - new_review = response["messages"][-1].content - state["draft_review"] = new_review - state["round"] += 1 - return state - -# Graph definition -from langgraph.graph import StateGraph - -graph = StateGraph(CodeReviewState) - -graph.add_node("draft_review", draft_review) -graph.add_node("reflect", reflect) -graph.add_node("rewrite", rewrite) - -graph.set_entry_point("draft_review") -graph.add_edge("draft_review", "reflect") - -graph.add_conditional_edges( - "reflect", - lambda x: "END" if x["verdict"] == "ok" else "rewrite" if x["round"] < x["max_rounds"] else "END", -) - -graph.add_edge("rewrite", "reflect") -app = graph.compile() - -# Demo function -async def demo(): - sample_code = """def sort_numbers(arr): - return sorted(arr)""" - init_state: CodeReviewState = { - "code": sample_code, - "draft_review": "", - "criteria_scores": {}, - "weakest_criterion": "", - "verdict": "", - "round": 0, - "max_rounds": 2, - } - result = await app.ainvoke(init_state) - print("--- Draft Review ---") - print(result["draft_review"]) - print("\n--- Scores ---") - print(result["criteria_scores"]) - print("\n--- Verdict ---") - print(result["verdict"]) - if result["verdict"] == "needs_revision": - print("\n--- Final Review After Rewrite ---") - print(result["draft_review"]) - print("\n--- Final Scores ---") - print(result["criteria_scores"]) +async def main(): + # Example user query + user_query = "What are the main causes of climate change?" + result = await agent.ainvoke( + {"messages": [HumanMessage(content=user_query)]}, + {"configurable": {"thread_id": "session-1"}}, + ) + # Print the final assistant message + print(result["messages"][-1].content) if __name__ == "__main__": - asyncio.run(demo()) + asyncio.run(main())