Use LangChain to refactor PaperAgent

[Copilot]

Refactor PaperAgent's AI system to use LangChain framework instead of custom agent implementation.

Changes

• Core Agent Architecture: Replaced custom MainAgent and CodeAgent with LangChain's agent framework using create_react_agent and AgentExecutor
• Tool System: Migrated custom tool registration to LangChain's @tool decorator pattern with structured input schemas
• Prompt Management: Converted inline prompts to LangChain ChatPromptTemplate with standardized agent scratchpad formatting
• LLM Integration: Unified model access through LangChain's ChatLiteLLM wrapper, maintaining existing LiteLLM multi-provider support
• Code Execution: Preserved existing Python code execution and visualization capabilities within new tool structure

Example

from langchain.agents import create_react_agent, AgentExecutor
from langchain_core.prompts import ChatPromptTemplate
from langchain_community.chat_models import ChatLiteLLM

# Create agent with LangChain
agent = create_react_agent(
    llm=ChatLiteLLM(model=model_name),
    tools=registered_tools,
    prompt=prompt_template
)

executor = AgentExecutor(
    agent=agent,
    tools=registered_tools,
    handle_parsing_errors=True
)

The refactor maintains backward compatibility with existing API endpoints and database models while providing a more maintainable foundation for future agent enhancements.

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Preview	Comments	Updated (UTC)
paper-agent	Ready	Preview	Comment	Nov 10, 2025 3:00pm