argus · alt + `
1:43 PM
it only opens its eye when you ask
argus is a screen-aware ai assistant that lives in your windows tray. press a hotkey, ask it anything about what you are looking at, or hand it the keyboard and let it finish the task for you.
100% free · mit licensed · windows 10/11 · zero telemetry
nothing is captured until you press the key
we all have the same models in a browser tab. almost nobody has one that can see the thing they are stuck on.
you already know the feeling. an error you cannot parse, a settings panel with forty switches, a video with no captions — and the only way to ask about it is to screenshot, crop, upload, and describe what you already had in front of you.
argus removes that whole trip. one hotkey, and the model is looking at the same pixels you are. ask, or say agent and let it drive.
no account, no server of mine in the middle, no telemetry. your key, your machine, your screen. it is early — try it and tell me what breaks.
two modes. one hotkey.
ask about anything you are looking at — then keep asking
Argus captures the screen once, sends it with your question, and keeps the thread going. Follow-ups resolve against what was already discussed, so you never have to re-explain yourself.
- Understands screenshots, not just text — errors, dashboards, videos, code
- Every conversation is saved as a resumable thread
- Works with a free Gemini API key
say agent, describe the task, watch it get done
Argus looks at the screen, decides one action, performs it, then looks again — the same loop a person uses, not a blind script. A pulsing frame stays on screen the entire time, so it never operates silently.
- Launches real apps directly — no typing into Windows Search, no stray browser tabs
- One action at a time, re-checked against the screen after every step
- Esc stops it instantly, from anywhere, no matter what has focus
your screen is the most personal data you have
Argus was built around one rule: nothing is captured until the hotkey is pressed, and nothing captured ever touches disk. Only text — never an image — is saved to your chat history.
- Screenshots live in RAM only, discarded once a request finishes
- No Argus servers. Your screen goes to the model provider you choose, and nowhere else
- Fully self-hostable — read every line, or fork it
the loop, exactly as it runs
flowchart LR
A["Press Alt + `"] --> B["Screenshot captured
(RAM only, never saved)"]
B --> C{What did you ask?}
C -->|"a question"| D["Talk Mode
vision model"]
C -->|"'agent ...'"| E["Agent Mode
model + tools"]
D --> F["Answer streams
into the bar"]
E --> G["One action:
launch app · click · type · scroll"]
G --> H["Action runs
on your real PC"]
H --> I{Task done?}
I -->|no| B
I -->|yes| J["Summary shown,
overlay disappears"]
every agent step re-captures the screen before deciding the next move — never a blind sequence
argus is free. the model is up to you.
there is no argus account and nothing to subscribe to. you bring an api key, and you pay whoever made the model — usually nothing at all.
argus
the app itself, forever
includes
- talk mode, unlimited
- agent mode, unlimited
- teach mode and workflows
- every line of the source
gemini free tier
best for trying it and for daily use
includes
- a key in about a minute
- vision, so it can see the screen
- roughly 20 requests a day
- stack more keys to raise it
paid api
best if you lean on it all day
includes
- no daily request ceiling
- bring gemini, openai or claude
- billed by them, never by argus
- swap models with
/aimodel
< building something with it? > argus is mit licensed, so fork it, rename it, ship it inside your own app. no permission needed — a link back is plenty.
fork itten minutes, and you have never opened a terminal before
There is no installer yet — you build it from the source, which is four commands. Every step below says what you should see when it worked, so you always know whether to carry on.
Install the two things Argus needs
Both are free, both are one-click installers, and you only ever do this once. Take the defaults on every screen.
- Node.js — click the big LTS button. This is what runs Argus.
- Git for Windows — this is what downloads the code. You can skip it; see step 1.
Then restart your computer. Installers add things to a list Windows only re-reads on startup, and skipping this is the single most common reason the next step fails.
Open a terminal, and check they are there
Press Win + R, type powershell, press Enter. A blue-black window opens. That is the terminal — you type commands into it and press Enter.
Worked if each prints a version number, like v22.14.0. If either says "not recognized", that thing did not install — or you have not restarted since.
Download the code
Paste both lines, one at a time. This puts Argus in your user folder.
Worked if the text before the cursor now ends in \argus>. That means you are inside the folder, which the next commands need.
No Git? Download the ZIP, right-click it, Extract All. Then in the extracted folder, hold Shift, right-click empty space, and choose Open PowerShell window here — you can skip straight to step 3.
Install and start it
The first command downloads what Argus is built from and takes a minute or two. It is noisy; that is normal.
Worked if a small bar appears in the middle of your screen. Argus lives in the system tray — the little icons by the clock. No app window opens, and nothing appears in the taskbar. That is not a mistake, it is the whole point.
Leave that terminal open. Closing it quits Argus. Minimise it instead.
Get a free key from Google
Argus does not have its own AI — you point it at one. Google's free tier needs no card and takes about a minute.
- Go to aistudio.google.com/apikey
- Sign in with any Google account
- Click Create API key
- Pick any project it offers, or let it make one
- Click Copy. The key is a long string starting
AIza…
Keep it private. It is a password to your Google AI account. Argus stores it on your machine and sends it nowhere but Google.
Paste the key into Argus
Press Alt + ` — that is the key above Tab, left of 1. The bar appears. Type /key, a space, then paste with Ctrl + V, and press Enter.
Worked if the bar replies "Gemini API key saved. Ask away."
Ask it something
Open anything at all — a web page, an error message, a spreadsheet. Press Alt + ` and type a question about what you are looking at.
That is everything. Type /help in the bar to see the rest, or read the full feature guide.
what you need
- Windows 10 or 11
- Node.js — free, one click
- A free Google Gemini key — no card
- About ten minutes
if something went wrong
- "npm is not recognized" — Node.js is not installed, or you have not restarted since installing it.
- Nothing appeared — Argus has no window. Look in the system tray by the clock, and press Alt + `.
- The hotkey does nothing — another app already owns it. Click the tray icon to open the bar, then type
/hotkey Control+Alt+Space. - "Every Gemini key is over quota" — the free tier is 20 requests a day. Add a second key from a different Google project with
/key; quota is counted per project. - It stopped after you closed the terminal — that terminal was running it. Run
npm run devagain.
prefer a real .exe?
Build a one-click installer for yourself, then you never need the terminal again:
It appears in the dist folder. Run it once and Argus installs like any other program.
everything is set in the bar
/tutorial— the guided tour, no key needed/help— list every command/aimodel— choose which AI answers/hotkey— rebind the shortcut/history— resume a past chat
four things keep that honest
A pulsing red frame covers the screen the entire time it has control. It can never operate silently.
Esc kills it immediately, from anywhere, whether or not Argus has focus.
Every task is capped at 14 actions, so a confused model can't grind on indefinitely.
It re-reads the screen after every action rather than firing a blind sequence.
frequently asked questions
what to know about argus, privacy, and getting started.
what is argus?
agent and it goes and does the task for you.i just installed it — where do i start?
/tutorial. It is a guided tour of every mode, read inside the bar, and it is the one thing that works before you have set an API key — so it answers "what did I just install?" at the moment you are actually asking. Enter turns the page; /tutorial exit leaves.is my data private?
is argus watching my screen all the time?
what can it actually do?
what is the difference between talk and agent?
which apps does it work with?
do i need windows?
is it really free?
what happens if i hit the free tier limit?
/key — quota is counted per project, and Argus rotates through the keys it has. Or point it at a paid key with /aimodel.can i fork it?
press a hotkey. ask about your screen.
free, open source, and running in about ten minutes.