Favicon of GPTeam

GPTeam

An MIT-licensed Python simulation of collaborating AI agents with separate memories. Runs locally and relies on legacy cloud-model integrations.

GPTeam is a self-hosted AI agent simulation for developers and researchers studying how agents communicate and work toward assigned goals. It runs locally in Python, while configured providers handle model calls through cloud APIs. The code uses the MIT license.

Each agent has its own memory and reflection process. Agents can talk to one another, collaborate on tasks, and work in parallel within a simulated world. Their locations matter: they move between places according to their tasks and where other agents are. This gives the project a focus on interactions between agents as well as the work each one performs.

You can change the agents and locations to explore different scenarios. GPTeam also records a summary of each agent's current state, so you can inspect what individual agents are doing while the simulation runs. Its memory and reflection approach draws on research into simulated agents.

The default model is GPT-4, with a GPT-3.5 Turbo alternative. Its Claude 1 integrations reference retired models and are no longer usable as documented. The GPT-4 and GPT-3.5 aliases are legacy choices with an announced October 23, 2026 shutdown. These API modes need an internet connection and provider credentials; running the simulation on your machine doesn't keep model requests local. GPTeam also supports Discord integration and a connection to the Window extension.

Similar to GPTeam