What Emu does
Emu does five things for you, all on this device. It watches and scores every AI action, holds destructive actions for your approval, stops secrets from leaving, masks personal information on the way out, and keeps a private log with a weekly recap.
1. Watches and scores every AI action
Everything your AI apps do is read by a small model that runs on your computer. A chat turn in ChatGPT. A tool call in Cursor. A command in Claude Code. A request to an MCP server. Emu scores it 0 to 100 for risk and records a short reason. The score stays on your machine.
This works across web chats, AI code editors, CLI tools, and MCP servers. You do not have to teach Emu where your AI apps live. It already knows the roughly 47 hosts on its list, and it intercepts them automatically once you install.
The on-device model is a small language model, about 800 MB. It does the watching, the scoring, and the Ask Emu pane. It does not send your prompts to anyone.
2. Holds destructive actions for your approval
When an AI is about to do something consequential, Emu pauses it and shows you a small prompt. Delete files. Force-push over a branch. Run a shell command. Drop a database. Emu steps in front of the dangerous move and asks: allow or block. If you are not around to answer, the destructive action does not happen.
This works across MCP servers, CLI tools, and AI code editors. The card slides in from the top-right of your screen, even when you are in another app, so the dangerous move waits for your call.
The shock moment, Emu steps in
You are always the one who decides. Emu never allows or blocks a risky action on its own. It just makes sure the dangerous ones stop and wait for you.
3. Stops secrets from leaving
If a request is about to send a password, API key, or credit card to an AI service, Emu stops it before it leaves your computer. You get three choices: send anyway, remove the secret and send, or cancel.
The default is to cancel. If you are not around to answer, the request does not go out. This is one of the two places Emu is allowed to be stricter than no guardian at all, because the alternative is a secret leaving.
4. Masks personal information on the way out
Phone numbers, email addresses, credit cards, Social Security numbers, API
keys, IBANs, and more get replaced with placeholders before they reach the
AI service. The AI service sees [PHONE_1234] and treats it as a value. It
responds normally. The original, unmasked version stays on your computer in
Emu's private log.
You pick how aggressive the masking is. Four presets cover most cases:
| Preset | What it does |
|---|---|
| Default | The everyday set. Phone, email, credit cards, SSNs, API keys, IBANs. |
| Healthcare | Default plus health-signal patterns. |
| Finance | Default plus finance-signal patterns. |
| Strict | Everything above plus tighter entity detection. |
Masking is the younger half of Emu. Name detection is off by default because the pattern detector is not reliable enough yet. Reply restore (swapping real values back into replies) is on the roadmap.
5. Keeps a private log and a weekly recap
Every AI interaction is recorded locally so you can search it later, get a weekly "your week in AI" summary, and ask Ask Emu questions about your own history. The notebook never leaves your computer.
Ask Emu is a right-side pane that opens with a keyboard shortcut. It runs the same on-device model that does the watching. You ask a question about your own history, it searches the private log, and it answers with citations. Every claim is one click from its evidence.
The raw traffic log is purged on a timer (the default is seven days). The brain's paraphrased summaries stay under a size-based storage budget, so the recap keeps working without your prompts piling up forever.
Where to go next
- Modes for the three ways Emu can treat an app (off, observe, guardian) and the full list of what Emu watches.
- Privacy for what stays on your computer, the one thing that leaves, and the uninstaller that wipes everything.
- Get started for the five-beat install.