for anyone running AI on their own machines

Your AI works hard. Is anyone watching it?

Every local model and agent you install can reach your files, keys and customer data. Nobody else keeps an eye on it for you, so Fleet does: one friendly table, and a message on your phone the moment something looks off.

Start free, 3 hosts See how it works

Latest checker v0.5.0 · 8 Oct 2026 · Open source · Made in Singapore

why this is on you

You installed it, so you are the one who can watch it

🔭

Vendors can't see your machine

They ship the software. They don't know which port you opened, which folder you shared, or which laptop got forgotten.

🧩

IT tools don't know what an agent is

Antivirus and cloud scanners miss a model server with no password, or an agent with shell access. To them it looks normal.

⚖️

The PDPA points at you

In Singapore the organisation must protect the personal data it holds, whatever tool touches it. "The AI did it" won't cover you. Showing you were watching helps.

🤖

Agents act on their own

They run commands and read pages without asking. You need to know what they can reach before something tricks them.

🔔 You don't expect the sofa maker to watch your living room. You fit a smoke alarm. Fleet is that alarm for your AI.

what you get

One table. Worst host first.

Each machine runs the free checker and reports a score out of 10. Sample data below. Your own hosts show up once you add the first one.

Host	OS	Score	What it found	Last seen
gpu-box-01	Ubuntu	3	Model server open to the whole network	2 min ago
dev-laptop-ana	macOS	6	Agent can read the whole home folder	11 min ago
office-mini	Windows	9	Nothing to fix	4 min ago
old-demo-vm	Ubuntu	–	Went quiet 3 days ago	3 days ago

Sample data, not real machines.

👀

See every agent

Which agents are on each machine, what they can reach, and which ones nobody has used in days.

💬

Alerts you can read

Telegram, Slack or webhook, in plain English: what's wrong, the risk, what to do. Plus a weekly email.

📴

Spot the quiet ones

A host that stops reporting gets flagged. You should know about a box that went dark.

👆

Fix it from your phone

Where a safe fix exists, the alert has a button for it, and an undo.

self-check

What are you answerable for today?

Tick what you run

A local model server (Ollama, LM Studio, vLLM, llama.cpp)A coding or desktop agent (Claude Code, Cursor, Codex, Cline...)MCP servers or tool pluginsAn agent that reads web pages, email or documentsAPI keys in environment files or shell historyMore than one machine doing any of the aboveI update these tools when I remember

Nothing ticked yet. Most teams tick three or more.

two minutes

Start with one machine

Run the checker, no account needed:

pipx install llm-host-guard
llm-host-guard

Sign in with your work email (magic link, no password) and create your organisation.
Copy the enrol key from Settings into the checker. The machine appears in your table. Repeat per machine.

Free for up to 3 hosts. More: fleet@homeauto.sg.

What leaves your machine

The score, the findings, the hostname and the names of detected agents. Not file contents, prompts or secret values. The checker is open source, so you can read exactly what it sends.

Your AI, your alarm.

Takes a few minutes. Stop any time by rotating the enrol key.

Get started free