Go-native · edge-ready · MIT

The AI agent that runs on a $10 chip.

A complete personal agent runtime in one Go binary: persistent memory, model routing, skills, schedules, and messaging—without hauling a data center onto your desk.

git clone https://github.com/operatoronline/operator-os.git
Get the Pi-in-an-hour kit → See the 3-step setup
<10MB idle RAM <1s cold start 5 architectures
● gateway online
arch/riscv64 ready
OPERATOR // EDGE RUNTIME
<10MB IDLE RAM
GO BINARYMIT/OSS

Memory, not marketing

Leave the runtime overhead behind.

Operator OS idles under 10MB. That changes where an always-on agent can live—and how little hardware it takes to keep one running.

99% smaller

Than the displayed 1.52GB OpenClaw reference. Against the bare Node.js baseline shown here, Operator uses over 74% less memory.

Operator OS <10 MB
Node.js ~39 MB
OpenClaw 1.52 GB

Reported idle/initial RSS; environments differ. Operator figure: project README. Node baseline: runtime benchmark. OpenClaw figure: community comparison.

Small binary, full stack

Everything the agent needs. Nothing it doesn’t.

01 / MODEL ROUTING

Bring your model. Switch without rebuilding.

Route across cloud and local providers from one configuration. Keep the runtime fixed while the intelligence layer changes.

OpenAIAnthropicGeminiGroqOllamaDeepSeek
02

Meet users where they are

Telegram, Discord, Slack, and WhatsApp channels are built into the gateway.

03

Memory that survives

Long-term memory and daily notes persist across sessions and reboots.

04

Skills, on disk

Add capabilities through a filesystem-based skills system you can inspect and version.

05

Time is a tool

Native cron scheduling turns prompts into repeatable background jobs.

Five architectures

Build once. Put intelligence anywhere.

From a Pi Zero-class board to a rack server, the same single-process design travels without a JavaScript runtime or Python environment.

release matrix
  • ✓x86_64desktop / server
  • ✓ARM64Pi / edge
  • ✓ARMv7legacy boards
  • ✓RISC-Vopen silicon
  • ✓LoongArch64native target
$10

Pi Zero-class

An always-on personal agent without always-on cloud compute.

0

Runtime baggage

A self-contained Go binary. No Node.js or Python runtime to manage.

1

Gateway daemon

One process coordinates channels, sessions, tools, memory, and schedules.

MIT

Open source

Inspect it, fork it, and deploy it on hardware you control.

From clone to online

Three moves. One running agent.

Onboard the workspace

Initialize the folders and configuration Operator uses for your agent.

operator onboard

Configure your keys

Choose a model provider and connect whichever messaging channels you need.

$EDITOR ~/.operator/config.json

Start the gateway

Bring the daemon online. Your configured channels route into the agent loop.

operator gateway

Pick your launch path

Build it yourself—or hand us the keys.

Start with the exact parts and commands, or get a production-minded setup tailored to your hardware, models, and channels.

THE DIY PATH
$49

Pi-in-an-hour kit

A focused deployment kit for getting Operator OS running on inexpensive edge hardware without losing a weekend to configuration.

Get the kit →

Operator OS is optimized for self-hosted, single-user deployments. Scope and final pricing are confirmed before work begins.

Command copied