Skip to main content

Private AI operators for your homelab

ArgoBox is the approval-first control plane between OpenClaw and your infrastructure.
Hermes routes. ArgoBox gates. You approve.

docker compose up -d

The canonical v1 stack

ArgoBox

Approval queue, audit trail, capability gating, and the operator UI.

Hermes

Operator-facing routing layer that decides what should execute, what stays read-only, and what needs approval.

OpenClaw

The agent runtime that asks ArgoBox for actions instead of receiving blind shell authority.

Home Assistant

The home automation substrate for the homelab-first wedge. ArgoBox wraps it instead of reinventing it.

How it works

Agent requests action

Your AI agent wants to restart a VM, stop a container, or change DNS. It calls ArgoBox instead of raw SSH.

ArgoBox validates

Capability check. Rate limit. Safety classification. Mutations go to the approval queue.

You approve

One click. Browser notification. Sound alert. Full audit trail of every action.

The first proof workflow

1. OpenClaw calls POST /api/agent/execute
2. ArgoBox classifies the request
3. Mutations enter the approval queue
4. Operator approves or rejects
5. Execution result is audited and returned
Read the operator loop

What you get

Approval Queue

Mutations require human approval. One-click approve/reject.

Audit Trail

Every action logged — who, what, when, why, result.

Rate Limiting

Per-agent quotas. Requests/min, mutations/hour, mutations/day.

Capability Gating

Agents get bounded capabilities instead of broad root-style trust.

Homelab Adapters

Connect Proxmox, Docker, Unraid, Home Assistant, and other approved targets through one control plane.

One Supported Install Path

Start with one documented Docker-based path before expecting framework-level breadth.

Get started with the supported path

git clone https://github.com/inovinlabs/argobox.git
cd argobox
cp .env.example .env
docker compose up -d
# Open http://localhost:4321/admin/setup
Read the full quickstart