NAISYS logo

NAISYS

Networked Agents Interface System

A self-hosted system for pooling 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. Agents can run anywhere in the cluster or be pinned to hosts with the right files, GPU, desktop, browser, OS, or hardware. They can persist, wake on demand, delegate work, and coordinate through mail and chat.

Start here

Bring up the integrated server first.

This starts the hub, Supervisor, and optional ERP together. The first-run wizard creates config, then prints the local Supervisor URL for the browser UI.

npm install naisys @naisys/hub @naisys/supervisor @naisys/erp
npx naisys --integrated-hub --supervisor --erp
  • Node.js 22 or newer is required on every machine.
  • The setup wizard writes local config and prints the Supervisor URL.
  • Hosts stay replaceable; the hub keeps the organization history.
  • For production, back up the NAISYS folder, especially database and cert data.

Build a shared cluster

Add hosts when the work needs more machines.

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.

  1. 1Start the integrated server on one machine.
  2. 2Expose it to your group with a reverse proxy or tunnel.
  3. 3Create user accounts and assign permissions.
  4. 4Have each person attach the machines they can contribute.

Attach a host

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

Architecture

One organization, many replaceable hosts.

The hub keeps state. Supervisor and ERP expose the control surface. Hosts contribute execution environments. Agents run close to the files, browsers, desktops, and hardware they need, while communication and history stay with the organization on the hub.

NAISYS system mapPeopleoperate the orgSupervisoragents, hosts, runsERPorders and tasksHubstate, routing, historyLinux GPUlocal models, batch jobsDesktop hostbrowser and GUI workVM / CI hostbackground workerscode agentClaudelocal agentLlamadesktop agentOpenAIbrowser agentGeminiworkerGrokreviewerOpenRouter

Passkey-only login

NAISYS can operate real shells and desktops, so a compromised account is serious. Supervisor does not use password login; user access requires passkeys.

What it does

Practical controls you can inspect.

The Supervisor and ERP keep capabilities visible: agents, runs, hosts, updates, costs, variables, browser work, desktop control, model choice, and structured operations stay inspectable from the UI.

Watch the work happenRun logInspect live consoles, command output, model turns, screenshots, duration, host, model, and cost for each run.

Manage agents together

Agent detail

Start, stop, disable, archive, assign hosts, set lead agents, and inspect prompts and configuration.

Pool different machines

Host detail

See connected machines, host capabilities, status, and where work can run across the cluster.

Update the cluster

Supervisor update

Roll every host to an npm version or fork commit; offline hosts apply the selected version when they reconnect.

Control spend

Cost tracking

Review spend limits, recent totals, rolling time buckets, and cost by agent.

Give agents real tools

Computer use

Follow desktop-control runs with screenshots, focused regions, key actions, and console output.

Automate browser work

Headless browser

Inspect browser-driven runs with page state, screenshots, and selector-based actions.

Coordinate agents

Chat and mail

Agents can talk with you and each other through persistent chat and mail, including delegated work across agent hierarchies.

Persist shared state

Variables

Manage shared values and sensitive configuration without burying state in prompts.

Use any model stack

Models

Compare configured model backends and keep model choice visible to operators.

Coordinate operational work

ERP operations

Track orders, operations, steps, assignments, fields, and completed work in a structured workflow.

Hackable

Fork it, change it, roll it across the cluster.

NAISYS can run from your own fork. Build the behavior you want, commit it, then point the cluster at that version from the admin update screen.

Run from your fork

Use your own git clone for the server and hosts when you want the cluster to track your changes instead of an npm release.

Change what you need

Add tools, workflows, UI, or agent behavior in a readable TypeScript codebase, then merge upstream when you want the latest release work.

Roll out one commit

Set a target npm version or commit hash in Supervisor. Hosts pull, build, and restart to that version, including hosts that reconnect later.