Build stateful agents with the popular LangChain framework. This template uses LangGraph to create cyclical, stateful agent runtimes that can manage complex reasoning flows and tool use within the ARC-AGI-3 environment.

​
LangGraph Templates

​
Multi-Modal

# Run the LangGraph agent with image support
uv run main.py --agent=langgraph --game=ls20

​
Text only

# Run the LangGraph agent (text-only)
uv run main.py --agent=langgraphtextonly --game=ls20

​
Random Agent

Created by Anusheel Bhushan (@anusheel), Burak Akkaya (@bakkaya-plank), Sophia Willows (@sophiabits) of Plank. 
# Run the LangGraph random agent
uv run main.py --agent=langgraphrandom --game=ls20

​
LangGraph Thinking Agent

A more advanced template than LangGraph Random Agent for people to build upon. Uses a variety of different tools, and has support for making long-term observations which persist between runs. Created by Anusheel Bhushan (@anusheel), Burak Akkaya (@bakkaya-plank), Sophia Willows (@sophiabits) of Plank. 
# Run the LangGraph thinking agent
uv run main.py --agent=langgraphthinking --game=ls20