Homelab-first v1 • approval-first agents

Run private AI operators on your homelab.
Approve every mutation.

ArgoBox is a private AI control plane that sits between AI tools and your infrastructure. The current public wedge is one hardened stack: ArgoBox, Hermes, Argonaut, and Home Assistant. No raw shell trust. No vague autonomy story. One canonical path first.

See the Product Read the Quickstart

0 Canonical Stack

0 Core Roles

0 Raw Shell Trust

0 Your Infra

argo@argobox:~

argo@argobox:~$ docker ps --format 'table {{.Names}}\t{{.Status}}\t{{.Ports}}'

NAMES STATUS PORTS

traefik Up 47 days 80/tcp, 443/tcp

portainer Up 47 days 9443/tcp

uptime-kuma Up 47 days 3001/tcp

homepage Up 47 days 3000/tcp

grafana Up 12 days 3100/tcp

prometheus Up 12 days 9090/tcp

jellyfin Up 30 days 8096/tcp

arr-stack Up 30 days 7878/tcp, 8989/tcp

Start With the Hardened Core

Product story first: what the stack is, how to install it, and how the approval-first trust boundary works.

What ArgoBox Is

See the homelab-first product framing: ArgoBox as the approval and audit layer, Hermes as routing, and Argonaut as the operator-facing assistant.

Product Control Plane Homelab First

Open

Canonical Quickstart

Install the supported v1 stack first. One documented path, one operator flow, one place to start before broader framework claims.

Install Docker Home Assistant Argonaut

Read

Approval-First AI

Read the public assistant story: scoped context, capability boundaries, auditability, and what ArgoBox still refuses to promise.

Threat Model Approval Queue Audit Trail

Review

Who Comes First?

ArgoBox is optimizing for homelabbers first, while keeping a clear path for developers and AI operators after the hardened core is proven.

Developers

The framework story is real, but it follows the hardened stack. Expect adapter seams, reference modules, and cleaner reuse after the core proves itself.

Framework Adapters Reference Modules

See the path

Homelabbers

This is the first wedge: a private AI control plane for Proxmox, Docker, Unraid, Home Assistant, and the rest of your self-hosted stack.

Primary Wedge Approval First Self-Hosted

Start here

AI Operators

The interesting question is not “can an agent act?” but “what controls survive hostile use?” ArgoBox leads with that trust boundary.

Argonaut Hermes Security

Review the model

Latest Transmissions

View Archive

Jun 4, 2026 ai

Stop Calling Agents Employees

The fastest way to build unsafe AI infrastructure is to pretend a model is a teammate with judgment. Treat it like an untrusted actuator with receipts.