LoopGPT is an open-source Python framework created by Fariz Rahman (farizrahman4u) that reimplements the Auto-GPT autonomous agent pattern in a cleaner, more modular package. It is installable via pip and designed to be both accessible and extensible for developers building autonomous AI agents. The framework is optimized for GPT-3.5 as well as GPT-4, and experimentally supports LLaMA and StableLM models, making it usable without a GPT-4 subscription. It requires no external vector database or vector store by default, though these are supported if desired, keeping setup friction low. A distinguishing feature compared to Auto-GPT is its human-in-the-loop design: when a user denies a command, the agent does not terminate but instead solicits feedback to correct its course. Full state serialization allows saving and reloading the complete agent state (configuration, conversation history, memory, and tool states) to a file or Python object without any external dependencies. LoopGPT follows a plug-and-play architecture: built-in tools cover web search and filesystem operations, and custom tools can be added with minimal code. Constraints can be attached to agent tasks, and a built-in math tool is included. The repository has approximately 1.4k-1.5k GitHub stars and 129-131 forks. Commit activity appears to have slowed significantly after 2023; the project is best described as minimally maintained as of 2025-2026. Key features: - Full agent state serialization (config, memory, tool states) to file or Python object - Human-in-the-loop feedback loop: agent course-corrects on user denial rather than terminating - Plug-and-play Pythonic API for adding custom tools and integrations - GPT-3.5 and GPT-4 support with experimental LLaMA and StableLM compatibility - No external vector database required by default - Built-in tools: web search, filesystem operations, math - Task-level constraints to scope agent behavior - Installable via pip as a standard Python package
Free and open-source (MIT license). Users supply their own OpenAI API key.