A device gets racked. A hostname changes. Someone's laptop joins a VLAN nobody remembers creating. The inventory you kept by hand is wrong again — usually right when you need it most. Pharos is a self-hosted directory for every machine, host, and person on your network. It stays current automatically. No spreadsheet to update by hand.

Try it without installing anything first:

sandbox
$ curl -sSL https://raw.githubusercontent.com/pharos-systems-llc/pharos/main/deploy/sandbox.yml -o sandbox.yml && \
podman-compose -f sandbox.yml up -d && \
rm sandbox.yml
Spins up a disposable sandbox; nothing to clean up but the one file it downloads.
Architecture

Three pieces do the actual work

Each one does a different job. Together, they're Pharos.

pharos-server

A read-optimized Rust backend daemon. It serves your inventory over RFC 2378, so every other piece has one place to ask.

pharos-console-web

The web console, for day-to-day lifecycle tracking. It also speaks Anthropic's Model Context Protocol — Pharos calls its browser-native transport WebMCP — so an AI agent can query your inventory directly instead of working from a stale export.

The data is schema-validated enough that an agent acting on it could do something like isolate an unrecognized MAC address to a restricted VLAN, not just flag it in a dashboard.

ph and mdb

Two fast CLI clients, for the lookups you'd rather run from a terminal than click through a UI.

A quick lookup, from the terminal

mdb skips the console entirely for a machine lookup like this one. Ask it for exactly the fields you want, and it answers right there in your shell:

mdb
$ mdb -H 'switch-core-02' return hostname status vlan last_seen_at
hostname: switch-core-02
status: online
vlan: 110
last_seen_at: 2026-10-02 16:29:45
Illustrative example — not a captured screenshot of a real lookup.

What shows up in the console

pharos-console-web turns that same lookup into something browsable — and makes it obvious the moment something drops off the network instead of quietly going stale. Here's that same switch alongside a second device, to show what a device that's gone offline looks like too:

switch-core-02

Online
VLAN
110
Last seen
Oct 2, 2026, 4:29 PM

sensor-garage-03

Offline
VLAN
40
Last seen
Sep 29, 2026, 8:14 AM

A simplified example of two device entries — not real captured data.

Discovery & presence

How a device actually gets in here

Two separate tools handle this — one for devices Pharos hasn't seen yet, one for devices it already has.

pharos-scan

Finds devices nobody's told Pharos about yet. An interactive run browses for SSH-reachable devices over mDNS. A scheduled --auto run does something different — an ARP-cache sweep, the real mechanism behind automatic rediscovery — so new devices turn up without anyone running a scan by hand. Either way, it classifies what it finds by the ports it sees open (a Proxmox host, a Plex server, a web server, a plain SSH box, or just "unknown" if nothing matches) and writes a basic record into Pharos.

pharos-pulse

Keeps an already-known device's record honest. Installed directly on that device, it sends one baseline inventory, then a heartbeat every hour — real hardware detail, not a guess from an open port, and a reliable online/offline signal the moment it stops checking in.

pharos-scan won't overwrite what pulse already knows: every record remembers which tool wrote it, and pulse's data always wins.

Where it runs

Pharos adapts to where you're actually running it.

Local dev

In-memory. Nothing to configure, nothing left behind when you stop it.

Homelab

File-based storage. Comfortable running in a Proxmox LXC or anywhere else you'd run a small self-hosted service.

Enterprise

LDAP-backed, for an environment that already has one.

AGPL-3.0

It's open source, with one distinction worth knowing before you deploy it: run it at home and you can modify it however you want, no strings attached.

Run it as a networked service for an enterprise or SMB, and you need to contribute changes back and keep attribution intact.