Fourteen-year-old developer Stanislaus Hees from Teydra Lab has released OpenAutark, an open-source desktop application designed for local multi-agent collaboration. The platform allows users to run AI agents locally, reducing reliance on cloud-based APIs and mitigating risks related to data privacy and service outages.
OpenAutark supports a hybrid model approach, enabling users to switch between local models—powered by tools like llama.cpp, Ollama, or vLLM—and external large language models. This architecture is designed to optimize costs by handling routine tasks locally while reserving expensive cloud API calls for complex operations.
The application features a manager agent that distributes tasks among specialized sub-agents, each with its own memory and role. It includes standardized protocols for tool integration, such as file system access and database interaction, while maintaining a human-in-the-loop control mechanism to ensure decision transparency.
The software is compatible with Windows, macOS, Linux, and Android. Currently in its early development phase, the platform's performance and the number of concurrent agents are limited by the available RAM on the host device.
