OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Multi-Agent Systems

Generativeagentscn

Generativeagentscn is a Python reconstruction and Chinese localization of the Generative Agents virtual-town simulation. It runs 25 LLM-driven agents in a simulated environment, supports local Ollama models and other OpenAI-compatible APIs, can resume interrupted runs, and provides browser-based replay plus Markdown timelines of agent activity and dialogue.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Generativeagentscn — OpenViST project profile in Multi-Agent Systems.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it for hands-on experiments with Chinese-language multi-agent interactions when you can run a Python project locally and configure the model backend yourself.

Who it is for

  • Researchers exploring simulated social behavior among multiple LLM-driven agents.
  • Developers adapting Chinese prompts or interaction patterns for a virtual-town simulation.
  • Users who want to inspect simulation activity through replay and saved timelines.

Documented uses

  • Run a named virtual-town simulation with configurable time steps.
  • Configure local Ollama models or another OpenAI-compatible model API.
  • Resume a run after it has stopped or been interrupted.
  • Generate replay data and review agent movement, timelines and dialogue.
Before you choose

Requirements & limits

The repository is public and licensed under Apache-2.0. The documented workflow runs locally and requires Python dependencies plus a configured model backend.

  • The project is a self-hosted Python research codebase; it is not a hosted, click-to-run simulation.
  • The README recommends Python 3.12 in a Conda environment.
  • Ollama supports local model use; other OpenAI-compatible providers require configuration of the provider, model, API key and base URL.
  • The project includes Chinese-localized prompts and documents support for Qwen3 and DeepSeek-R1.
  • Replay is a separate preparation and service stage; the project also saves activity and dialogue in Markdown.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up and run a virtual-town simulation

Prepare the Python project and launch a named simulation run.

Before you start

  • Python 3.12; the README suggests using a Conda environment.
  • A configured local Ollama model and embedding model, or an OpenAI-compatible API with its provider, model, API key and base URL.
  • The project repository and its Python dependencies.

Steps

  1. Clone the repository and enter its directory.

  2. Configure generative_agents/data/config.json for Ollama or another OpenAI-compatible API.

  3. Create and activate the suggested Conda environment, then install dependencies with the command below.

    Documented command
    pip install -r requirements.txt

  4. From the generative_agents directory, run the command below with a unique name, start time, step count and stride.

    Documented command
    python start.py

Expected outcome

A named simulation run is started; its name can be used later for replay.

Guide limitations

  • Local Ollama use requires installing Ollama, downloading compatible models and matching the configured model and base URL.
  • The README's command-line example uses a specific start time and simulation parameters; choose values appropriate to the intended run.
Documentation 1 ↗
Source-based practical guide

Prepare and view a simulation replay

Create replay files from a named run and open its activity in the browser interface.

Before you start

  • A completed or otherwise available named simulation run.
  • The project environment with its Python dependencies installed.

Steps

  1. From the generative_agents directory, run the command below followed by the run name.

    Documented command
    python compress.py --name

  2. Run the command below to start the replay service.

    Documented command
    python replay.py

  3. Open the local replay page with the same run name in its name query parameter.

Expected outcome

The replay data includes movement information, while `simulation.md` presents agent activity and dialogue as a timeline; the browser page shows activity over time.

Guide limitations

  • The README gives http://127.0.0.1:5000/ as the local replay address and says the view is navigated with the keyboard arrow keys.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
x-glacier/GenerativeAgentsCN · Official homepage title · retained in its source language
How we review sourcesSuggest a correction

Related options

wounderland — The README identifies wounderland as the refactored Generative Agents project on which Generativeagentscn is based.

Generative Agents — The original project that Generativeagentscn reconstructs and localizes in Chinese.