NAISYS

NPM | Website | Discord

NAISYS is a self-hosted system for a group of people to pool machines, model accounts, and work into one shared organization of humans and AI agents.

Run a local server, open the browser UI, create agents, invite people, and let everyone add the machines they can contribute. A laptop can handle chat or code agents, a GPU box can serve local models, a Windows desktop can run computer-use tasks, and a VM can take always-on background work. Agents can run anywhere in the cluster or be pinned to specific hosts.

Start here

Requires Node.js 22+ on every machine (node -v).

npm install naisys @naisys/hub @naisys/supervisor @naisys/erp
npx naisys --integrated-hub --supervisor --erp

The first-run wizard creates the local config, then NAISYS prints the local Supervisor URL. Open that URL and use the GUI from there: create agents, add more hosts, invite users, assign permissions, inspect consoles, review runs, send mail, and track cost.

For a production install that auto-restarts, see PM2 deployment.

Build a shared cluster

1. Start the integrated server on one machine.
2. Make it reachable by the group with a reverse proxy or a tunnel like ngrok.
3. Create user accounts for the people who should help manage the organization.
4. For each remote machine, add the host in the supervisor (Hosts → Add Host, copy the generated access key from the one-time modal), then on that machine run these commands which will as for the HOST_ACCESS_KEY:

npm install naisys
npx naisys --hub=https://<server>/hub

The hub keeps logs, mail, cost history, variables, and agent state. Hosts stay replaceable: if a machine disconnects, the organization still has its history and can run work somewhere else.

What you can do

• Manage people and agents together - users, permissions, API keys, human-managed agents, and LLM-backed agents in one system
• Pool different machines - desktops, laptops, VMs, GPU boxes, Windows/Mac/Linux hosts, and restricted hosts for explicitly assigned agents
• Assign work by host - run agents anywhere in the cluster or pin them to machines with the right tools, OS, GPU, browser, or desktop session
• Watch the work happen - live consoles, run logs, screenshots, mail, chat, host status, and command-level history from the browser
• Control spend - per-agent spend limits, rolling windows, model pricing, cost charts, and suspended status when an agent hits its cap
• Use any model stack - OpenAI, Anthropic, Google, OpenRouter, Grok, and local LLMs
• Give agents real tools - shell, headless browser, desktop computer use, image generation, audio, inter-agent mail, and agent-facing REST APIs
• Coordinate operational work - optional ERP for orders, operations, steps, assignments, labor tickets, inventory, and agent-readable workflows

Packages

NAISYS is split into four packages, each with its own README:

• naisys — the agent runner
• @naisys/hub — central server for persistence and cross-machine communication
• @naisys/supervisor — web UI for monitoring and managing agents
• @naisys/erp — AI-optimized order/task management system

Data, logs, and backups

The setup wizard writes .env; NAISYS_FOLDER in that file controls where persistent state lives. Databases are stored under NAISYS_FOLDER/database/, logs under NAISYS_FOLDER/logs/, and bootstrap access keys for the integrated naisys + supervisor hosts under NAISYS_FOLDER/cert/ (remote hosts' keys are not stored here — they live in each remote machine's HOST_ACCESS_KEY).

For production, back up the full NAISYS_FOLDER. At minimum, preserve database/ and cert/ together so restored hosts can keep their data and the integrated processes can reauthenticate. Remote hosts re-key through the supervisor UI if needed.

Details

• PM2 deployment for production servers and hosts
• XFCE / VNC host setup for desktop computer-use agents
• Release and versioning
• Architecture overview — how the packages fit together
• HATEOAS API design — how the supervisor and ERP expose AI-first REST APIs
• Computer use — coord spaces, focus, and per-vendor contracts for desktop control
• Database design — Prisma strategy, the three schemas, migrations
• Agent config options and supported models

Other design docs under docs/ cover multi-machine operation, mail, web auth, supervisor users, port strategy, hub security, and attachments.

Running from source

git clone https://github.com/swax/NAISYS.git
cd NAISYS
npm install && npm run build
node apps/naisys/dist/naisys.js --integrated-hub --supervisor --erp

Requirements

• Node.js 22 or higher
• A supported shell — NAISYS proxies one to the LLM:
  • Linux (bash/sh)
  • Windows (PowerShell/WSL)
  • macOS (bash/zsh)
• Optional, per feature:
  • ns-lynx (text browser): lynx — apt install lynx or equivalent
  • ns-browser (headless Chromium): npm install playwright-core && npx playwright install chromium
  • ns-pty (TTY commands like sudo/ssh/passwd): Linux only
  • ns-desktop (GUI computer use) — supported per platform:
    • Windows: Works out of the box
    • macOS (preliminary): cliclick + screencapture
    • Linux X11: xdotool + scrot (or gnome-screenshot / import)
    • Linux Wayland: ydotool + grim
    • See the XFCE / VNC host setup guide for a recommended headless Linux setup

Troubleshooting

• Check node -v; every machine needs Node.js 22 or newer.
• If a server port is busy, set SERVER_PORT in .env and restart.
• If a host cannot connect to a remote hub, verify the --hub URL includes /hub and that HOST_ACCESS_KEY in the client's .env matches the key shown when the host was created (or rotated) in the supervisor.
• If ns-browser or ns-desktop fails, install the optional browser/desktop dependencies listed above.
• Check NAISYS_FOLDER/logs/ for hub, supervisor, ERP, and runner logs.

License

MIT