AOS-CE — Open Agent Operating System

Introduction

AOS Community Edition is an open-source operating system layer for AI agents. It provides process-level isolation, resource scheduling, and lifecycle management for autonomous agents, treating each agent as a managed process with defined permissions, resource limits, and communication channels.

What AOS-CE Does

• Manages agent lifecycles from spawn to termination with process-level isolation
• Schedules agent workloads across available compute resources with priority queues
• Enforces resource limits (CPU, memory, network, filesystem access) per agent
• Provides inter-agent communication channels for multi-agent coordination
• Logs all agent actions and tool calls for auditing and replay

Architecture Overview

AOS-CE is written in Rust and runs as a lightweight daemon on Linux. It uses OS-level namespaces and cgroups to sandbox each agent in its own isolated environment. A central scheduler assigns agent tasks to worker slots based on priority and resource availability. Agents communicate through typed message channels managed by the runtime, preventing direct memory sharing. The audit subsystem captures a structured log of every tool call and state transition.

Self-Hosting & Configuration

• Requires Linux with cgroup v2 support; single-binary installation via shell script
• Configure agent templates with resource limits and permitted tool sets in YAML
• Set up workspace directories that agents can access for file I/O
• Define scheduling policies (FIFO, priority, or round-robin) in the runtime config
• Enable the built-in web dashboard for monitoring agent status and resource usage

Key Features

• Rust-based runtime with minimal memory footprint and fast startup times
• OS-level sandboxing using namespaces and cgroups for agent isolation
• Structured audit log of all agent actions for compliance and debugging
• Inter-agent messaging with typed channels and backpressure handling
• Web dashboard for real-time monitoring of agent health and resource consumption

Comparison with Similar Tools

• Docker — general container runtime; broader ecosystem but not purpose-built for agent lifecycle management
• E2B — cloud sandbox for AI code execution; managed service rather than a self-hosted OS layer
• OpenSandbox — AI agent sandbox; similar isolation goals but AOS-CE adds scheduling and multi-agent coordination
• Kubernetes — orchestration platform; more powerful but far heavier than AOS-CE for single-node agent workloads

FAQ

Q: Does AOS-CE replace Docker? A: No. It complements container runtimes by adding agent-aware scheduling, lifecycle management, and inter-agent communication on top of OS-level isolation primitives.

Q: What AI agents does it support? A: Any agent that runs as a CLI process. It is agent-framework agnostic and works with coding agents, research agents, and custom agent implementations.

Q: Can it run on macOS or Windows? A: Currently Linux only, as it relies on cgroup v2 and Linux namespaces for sandboxing. WSL2 on Windows may work for development.

Q: Is it production-ready? A: AOS-CE is under active development. It is suitable for development and staging environments; evaluate carefully for production critical workloads.

Sources

• https://github.com/unicity-aos/aos-ce