AIOS is an open-source operating system layer for LLM-based agents, developed by the AGI Research group at Rutgers University. The project has collected about 6,500 stars on GitHub and remained under active development in 2026.
AIOS treats agents the way an operating system treats applications. Its kernel handles the unglamorous problems that appear as soon as many agents run at once: scheduling, context switching, memory and storage management, tool management, and access control. On top of the kernel sits the Cerebrum SDK, the interface developers use to build agents. AIOS runs in a local kernel mode on your own machine or against a remote kernel, ships both a Web UI and a terminal UI, and supports computer-use agents that operate a virtual machine through MCP tools. The underlying research was accepted at COLM 2025. The idea borrows directly from operating systems research: just as processes should not manage their own memory, agents should not each reinvent scheduling and tool plumbing. By moving those concerns into a shared kernel, AIOS aims to make large agent systems more reliable, more observable, and easier to scale than a pile of independent scripts.
AIOS is free and open source, published as a research project with no hosted cloud or paid tier. You supply your own model API keys and infrastructure.
Agent researchers and platform builders who are outgrowing single-agent scripts and need OS-level structure for running many agents at once.