agiresearch/aios

★ 5,676PythonAudience · researcherComplexity · 5/5Setup · hard

Mindmap

mindmap
  root((AIOS))
    What it does
      Multi-agent kernel
      Agent scheduling
      Memory management
    Architecture
      AI kernel
      Cerebrum SDK
      Tool management
    Supported models
      OpenAI
      Anthropic
      Hugging Face
      Ollama
    Use cases
      Research
      Agent frameworks
      Computer-use agents
    Audience
      Researchers
      ML developers

mindmap root((AIOS)) What it does Multi-agent kernel Agent scheduling Memory management Architecture AI kernel Cerebrum SDK Tool management Supported models OpenAI Anthropic Hugging Face Ollama Use cases Research Agent frameworks Computer-use agents Audience Researchers ML developers

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Run multiple AI agents in parallel, one searching the web, one writing a file, one managing storage, without conflicts

USE CASE 2

Build a custom agent using the Cerebrum SDK that runs on the AIOS kernel with shared memory and tool access

USE CASE 3

Experiment with computer-use agents that control a sandboxed virtual machine to perform tasks like a human

USE CASE 4

Research multi-agent scheduling and coordination by extending or modifying the AIOS kernel

Tech stack

PythonOpenAI APIAnthropic APIHugging FaceOllamaFastAPI

Getting it running

Difficulty · hard Time to first run · 1h+

Requires installing both the AIOS kernel and the separate Cerebrum SDK, plus an LLM API key or local Ollama instance.

Open-source research project from Rutgers University, specific license terms are in the repository.

In plain English

AIOS is a research project from Rutgers University that treats a large language model (an AI system like the ones that power chatbots) as the core of an operating system rather than just a tool you call once and get a response from. The idea is that if you want to run multiple AI agents at the same time (each doing something different: one searching the web, one writing a document, one managing files), you run into the same kinds of problems that a computer operating system solves for regular programs: how do you schedule them, how do you share memory between them, how do you manage their access to tools and storage without them stepping on each other? AIOS builds an "AI kernel" that sits between the AI agents and the underlying computer, handling those coordination problems. Agents built on top of AIOS can call tools like web search or code execution, store and retrieve memory, and run alongside other agents, with the kernel managing how their requests are queued and processed. The system supports both a web interface and a terminal interface. The project has two main parts. The kernel (this repository) manages the runtime: scheduling requests to the language model, handling memory and storage, managing tools, and providing a consistent interface for agents to use. The SDK (a separate repository called Cerebrum) is what developers use to build and run agents on top of that kernel. AIOS supports many popular language model providers including OpenAI, Anthropic, Google, and open-source models through Hugging Face and Ollama. It also works with agent frameworks that other researchers have already built. A recent addition is support for computer-use agents, meaning agents that can control a virtual machine to perform tasks on a computer like a human would, using a sandboxed environment to keep that safe. This is an academic research project that has produced several peer-reviewed papers. It is designed for researchers and developers building multi-agent systems, not for general end users looking for a ready-made AI assistant.

Copy-paste prompts

Prompt 1

Using AIOS and the Cerebrum SDK, create an agent that searches the web for the latest news on a topic and stores a plain-English summary in persistent memory.

Prompt 2

Set up AIOS locally with Ollama as the LLM backend and run two agents in parallel, one that writes a file and one that reads it, to see how the kernel handles concurrent access.

Prompt 3

Configure AIOS to use the Anthropic API and run the sample multi-agent demo from the README.

Prompt 4

Build an AIOS agent that uses a computer-use tool to open a browser in a sandboxed environment and fill in a form.

Prompt 5

How does AIOS schedule concurrent LLM requests from multiple agents? Walk me through the kernel's request queue with a code example.

Open on GitHub → Explain another repo

← agiresearch on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.