Prerequisites¶
This document will introduce the software and hardware environment and dependencies required before using AgentSociety.
Hardware Environment¶
AgentSociety has no special requirements for hardware environment, but it is recommended to use a computer with certain CPU performance and sufficient memory to avoid slow operation or memory shortage issues.
Recommended configuration is no less than:
CPU: 4 cores
Memory: 16GB
Software Environment¶
AgentSociety is developed using the Python language and requires Python 3.11 or higher version to be installed.
Meanwhile, due to the support situation of the parallel computing framework ray that AgentSociety depends on, the supported systems and architectures of AgentSociety include:
Linux X86_64
MacOS ARM
Note
Support for Windows by ray is experimental, so AgentSociety does not support Windows systems. If needed, AgentSociety can be run using Windows Subsystem for Linux (WSL) 2.
Large Model API¶
AgentSociety uses large models to complete agent reasoning, so at least one large model API needs to be configured.
AgentSociety supports the following large model API platforms and any OpenAI API compatible large model APIs:
ZhipuAI: Provides completely free large model
GLM-4.5-FlashvLLM: The vLLM framework can be privately deployed with OpenAI API compatible large model services. Note that the tool calling capability must be enabled.
Any OpenAI API compatible large model API: Configured through fields such as
base_urlandapi_keyin the large model related configuration
Additional Notes¶
Tip
Based on user experience, AgentSociety version 1.5 has significantly reduced the dependent software, removing dependencies on software such as MLflow and Redis, making it more convenient for users to use AgentSociety.
The database defaults to file-based storage using SQLite to avoid software dependencies. Users can still configure a PostgreSQL database through the db field in the configuration file to share data with team members.