Skip to content

Appliance (Collector)

The Acutis appliance — internally "the collector" — extends Acutis from individual endpoints to the whole network. Where the Acutis Go agent answers "is it this machine or the network?" from a single laptop, the appliance watches the network itself — switches, firewalls, and access points — and feeds it all into your fleet dashboard, The Floor.

It also runs its own local AI (the "smart collector"): an on-box model that reads the live state of your gear, explains what it sees in plain language, and — on paid tiers, with your approval — helps push the fix.

What it is

A small, always-on Ubuntu device that runs the Acutis collector. You can run it two ways:

  • Mini-PC — a compact fanless box (for example a Raspberry Pi 5 or an Intel mini-PC) plugged into your network. Good for a permanent on-site install.
  • OVA / VM — a virtual appliance you import into VMware, Proxmox, or any hypervisor. Good when you already run virtualization on-site.

Either way the software is identical. It sits on your management network, discovers what's there, polls the devices you onboard, and ships the results to your backend over HTTPS.

How it fits together

Endpoints ──(Acutis Go agent)──┐
                               ├──▶  Backend  ──▶  The Floor (cloud dashboard)
Switches / Firewall / APs ─(appliance)─┘
  • The Go agent reports the endpoint's-eye view (path, DNS, Wi-Fi, machine health) up to the cloud.
  • The appliance reports the infrastructure's-eye view (ports, sessions, routing, topology) up to the same cloud.
  • The Floor stitches both into one live picture of your site, so a problem can be pinned to the exact device — wired or wireless, machine or network.

Go agent vs appliance — the key difference

The Go agent is observe-only by design: a roaming laptop is never trusted to hold device passwords. The appliance is credentialed: it can read device configuration and — only on Pro and above, only after you approve each change — stage and deploy configuration to your gear. The two are deliberately different trust levels.

What it monitors

  • Switches — port status, errors (CRC), PoE draw, throughput, VLANs, spanning-tree state, MAC/forwarding tables, LLDP/CDP neighbors (the basis of topology).
  • Firewalls — interface and session health, CPU/memory, routing, top talkers (e.g. Palo Alto PA-series via the XML API).
  • Access points / Wi-Fi — radio metrics and connected-client health.
  • Servers & hypervisors — Proxmox, ESXi, XCP-ng, Hyper-V resource state.
  • Everything else on the wire — an unlimited, read-only nmap discovery of the subnet, so you see hosts even before you onboard them.

The local AI ("smart collector")

During first boot the appliance installs a small local language model (via Ollama) that runs entirely on the box — no device data leaves your building for it to work. Each poll cycle it turns the real telemetry into a short "field brief" and a triage flag, which it passes up to the cloud as pre-digested context for the senior AI assistant ("Carlo"). See The AI smart collector.

In this section

  • Requirements — hardware / VM specs, network placement, and outbound connectivity.
  • Install — the zero-touch first-boot flow, step by step.
  • Provisioning — provision.conf, the collector keys, and TLS pinning.
  • Onboarding devices — adding switches, firewalls, and APs; supported vendors; the credential vault.
  • The AI smart collector — read-only diagnosis vs. approved config changes.
  • Device limits by tier — managed vs. discovered devices, and the real per-tier numbers.
  • Troubleshooting — concrete checks when something isn't reporting.
  • Security — on-prem credentials, the zero-knowledge vault, and what leaves the building.

Getting an appliance

The appliance is available on every plan — Free included (2 devices, observe-only). You provision it from your account: create a site token, write it to the install media, and boot. No SSH required. Start at Requirements, then Install.