IntroductionHave you ever thought about how AI can not only chat and draw but also socialize freely in a virtual world and host parties? The MetaGPT Stanford Town module takes you into an AI virtual world full of infinite possibilities! This article will guide you step by step on how to quickly set up and run this amazing AI town, allowing your AI residents to explore, collaborate, develop friendships, and even host a grand lunch party!
1. Rapid Setup in 5 Minutes
1. Environment SetupFirst, clone the frontend repository and create a Python environment:
git clone https://github.com/crcresearch/agentic_collab
conda create -n env39 python=3.9
conda activate env39
pip install -r requirement.txt
2. Key Path ConfigurationTo avoid data confusion in the frontend, modify the storage path in <span>run_st_game.py</span>:
STORAGE_PATH = EXAMPLE_PATH.joinpath(r"D:\stanford_town\agentic_collab-dev\environment\frontend_server\storage")
2. Secrets to Start Dual Services
1. Start Backend EngineNavigate to <span>examples/stanford_town</span> directory and run the following command:
python run_st_game.py "Host a open lunch party at 13:00 pm" "base_the_ville_isabella_maria_klaus" "test_sim" --temp_storage_path "temp_storage"
Parameter Explanation
-
<span>idea</span>: The first AI’s “inner voice”, this example triggers a lunch party -
<span>fork_sim_code</span>: Select a preset resident relationship template -
<span>sim_code</span>: The save folder for this simulation
2. Frontend VisualizationNavigate to <span>environment/frontend_server</span> directory and start the frontend service:
python manage.py runserver
Access http://localhost:8000/simulator_home to observe AI dynamics in real-time.
3. Environment Interaction Practice
1. Map Initialization
from metagpt.environment.stanford_town import *
env = StanfordTownExtEnv(maze_asset_path="D:/path/to/MAZE_ASSET_PATH")
2. Intelligent Observation Mode
# View details of coordinates (72,14)
bed_info = env.observe(
EnvObsParams(obs_type=EnvObsType.GET_TITLE, coord=(72, 14))
) # Output: ["Isabella's master bedroom bed", "Sleeping"]
# Scan the environment within a radius of 10 tiles
surroundings = env.observe(
EnvObsParams(obs_type=EnvObsType.TILE_NBR, coord=(72,14), vision_radius=10)
)
3. Event Control Demonstration
# Remove event at specified location
action = EnvAction(
action_type=EnvActionType.RM_TITLE_SUB_EVENT,
coord=(72,14),
subject="Isabella Rodriguez"
)
obs, _, _, _, info = env.step(action) # Isabella's sleeping event is removed
4. Pitfall Guide
1. Fatal Error FixIf you encounter <span>TypeError: 'coroutine' object is not subscriptable</span>:
# Before (Error)
access_tile = env.observe(params)
# After (Correct)
access_tile = await env.observe(params) # Add await keyword
2. API Key Security Configuration
# Replace with your key in const.py
OpenAI(api_key="sk-your-key-here", base_url="http://127.0.0.1:5000/v1")
3. The original version has bugs, use the version below
git clone https://github.com/crcresearch/agentic_collab
conda create -n env39 python=3.9
conda activate env39
pip install -r requirement.txt
cd environment/frontend_server
python3 manage.py runserver
4. Backend Data Points to Frontend
# Modify run_st_game.py, set frontend cache path temp_storage
STORAGE_PATH = EXAMPLE_PATH.joinpath(r"D:\stanford_town\agentic_collab-dev\environment\frontend_server\storage")
# Modify metagpt\ext\stanford_town\utils\const.py, set STORAGE_PATH
STORAGE_PATH = Path(r"D:\a\wow-agent\storage")
embedding = (
# OpenAI(api_key=config.llm.api_key).embeddings.create(input=[text], model=model).data[0].embedding
OpenAI(api_key="",base_url="http://127.0.0.1:5000/v1").embeddings.create(input=[text], model="bge-m3:latest").data[0].embedding
)
5. Run Errors
Run error: File "metagpt\ext\stanford_town\roles\st_role.py", line 251, in observe for tile in nearby_tiles: TypeError: 'coroutine' object is not iterable
Reason: nearby_tiles is a coroutine object, not iterable. Use the await keyword when calling the observe method to wait for the coroutine to complete.
Modify: Add await to lines 126, 243, 252, 259, 557, 572, 581, e.g., nearby_tiles = await self.rc.env.observe
Run error: File "metagpt\ext\stanford_town\roles\st_role.py", line 243, in observe
File "metagpt\environment\base_env.py", line 80, in observe return res
UnboundLocalError: local variable 'res' referenced before assignment
Reason: In some cases, the variable res may not have been assigned. To fix this issue, initialize the res variable at the beginning of the function. Please add res = None at the start of the observe method.
Run error: File "metagpt\ext\stanford_town\roles\st_role.py", line 250, in observe
TypeError: 'NoneType' object is not iterable
Reason: nearby_tiles = await self.rc.env.observe returned NoneType.
It seems like the map did not load correctly, ignore it for now.
Run error: File "metagpt\environment\stanford_town\stanford_town_ext_env.py", line 223, in reset
super().reset(seed=seed, options=options) AttributeError: 'super' object has no attribute 'reset'.
Modify: Comment out this line.
Run error: File "metagpt\ext\stanford_town\roles\st_role.py", line 251, in observe for tile in nearby_tiles: TypeError: 'NoneType' object is not iterable.
Reason: nearby_tiles is None.
Run error: File "metagpt\ext\stanford_town\actions\gen_action_details.py"
ExtEnv.observe() got an unexpected keyword argument 'obs_params'
Modify: Remove the obs_params parameter
access_tile = await role.rc.env.observe(
EnvObsParams(obs_type=EnvObsType.GET_TITLE, coord=role.scratch.curr_tile)
#obs_params=EnvObsParams(obs_type=EnvObsType.GET_TITLE, coord=role.scratch.curr_tile)
)
5. Advanced Techniques
-
Memory Inheritance: Copy the <span>generative_agents/storage</span>directory from the original town to the MetaGPT path to achieve historical memory loading -
Multi-Agent Collaboration: Create AI residents with different personalities by modifying <span>base_the_ville_*</span>template parameters -
Event Chain Triggering: Use <span>EnvAction</span>with<span>ADD_EVENT</span>type to create a continuous event stream
6. Running Screenshots
ConclusionWith the modular design of MetaGPT, we can focus on the core logic to build complex AI social systems. Start this amazing town on your computer now and observe how AI residents create unexpected social stories! The 11-day AI Agent learning series has come to a celebratory end!!!
Learn AI Agent Development in 11 Days: Day T+5 – Implement Werewolf Game with MetaGPT: Multi-Agent Collaboration Practice
Learn AI Agent Development in 11 Days: Day T+4 – Write Poetry with Multi-Agent System: When Students Meet Teachers
Learn AI Agent Development in 11 Days: Day T+3 – Tutorial Agent: The Magic Tool for One-Click Tutorial Document Generation
Learn AI Agent Development in 11 Days: Day T+2 – How to Create Your Exclusive Agent with MetaGPT: Building SimpleCoder from Scratch
Learn AI Agent Development in 11 Days: Day T+1 – Easy Start with MetaGPT
Learn AI Agent Development in 11 Days: Day 11 – MetaGPT Agents and Multi-Agent Systems: Constructing a “Human” Society in a Digital World
Learn AI Agent Development in 11 Days: Day 10 – Build an Intelligent Question-Answering Assistant with Zigent Framework, Freeing Teachers’ Hands!
Learn AI Agent Development in 11 Days: Day 9 – Create Your Exclusive Tutorial Writing Agent with Zigent Framework
Learn AI Agent Development in 11 Days: Day 9 – The “Big Data Audit Lesson Plan” Written by the Agent
Learn AI Agent Development in 11 Days: Day 9 – The “Artificial Intelligence and Audit Lesson Plan” Written by the Agent
Learn AI Agent Development in 11 Days: Day 8 – Implement Philosopher Multi-Agent: A Philosophical Dialogue Across Time and Space with Zigent
Learn AI Agent Development in 11 Days: Day 7 – Getting Started with Zigent: Build Your Intelligent Search Agent.
Learn AI Agent Development in 11 Days: Day 6 – Agent with Web Search Capability.
Learn AI Agent Development in 11 Days: Day 5 – RAG Access to Agent.
Learn AI Agent Development in 11 Days: Day 4 – Create Agent with Llama-index.
Learn AI Agent Development in 11 Days: Day 3 – Implement Grading Agent.
Learn AI Agent Development in 11 Days: Day 2 – Handcraft a Basic Agent.
Learn AI Agent Development in 11 Days: Day 1 – Selecting Agent Framework – wow-agent.
Learn AI Agent Development in 11 Days: Day 0 – What is an Agent.