v1.0.0 · macOS + Windows
A private AI that floats over everything.
Gupt AI is a private desktop overlay for macOS and Windows. Bring your own API key — OpenAI, Claude, Gemini, Groq, or Mistral — and your prompts go straight to the model. Or let us provide the key and skip the setup entirely.
Plans from ₹599/mo · macOS & Windows · cancel anytime
Walk me through designing a rate limiter.
Start with the policy. Two solid choices:
- •Token bucket — smooth bursts, refill at a fixed rate.
- •Sliding window — stricter, more memory.
function allow(key) {
const b = bucket(key)
b.refill() // add tokens since last call
return b.take(1)
}Add a per-key TTL so idle buckets expire
// how it works
Three steps. No account.
No sign-up, no servers in the middle. It's your machine talking to the model you chose.
Download & open the overlay
Grab the build for macOS (.dmg) or Windows. It launches as a small always-on-top window.
Paste your own API key
Pick a provider and model in Settings. Your key is stored locally on your machine — the picker auto-saves as you switch.
Summon it over any app
Press ⌘⇧H to show or hide it anywhere. Ask a question, drop in a screenshot, or let it transcribe what you hear.
// where your data actually goes
On bring-your-own-key, nothing sits in between
Add your own provider key and your prompts go straight to the model — we never see them, your key, your audio, or your files. (Prefer no setup? Managed plans route through our secure, metered proxy instead.)
Gupt AI app
the overlay
Conversations, settings & knowledge files
electron-store, on your disk
Your API key
stored locally
Stored on disk · never uploaded by the app
AI provider
The one you chose
Called directly with your key. You pay the provider for usage — Gupt AI takes nothing and adds nothing in the middle.
- Receives only your prompt + key
- Saved chats, files & audio stay on your device
On BYOK there's no Gupt AI server in the path — it's direct, not proxied by us.
// bring your own model
Bring the key you already have
Five providers, many models. Switch between them from a picker that auto-saves — no restart, no re-paste. Or skip the setup and use ours on a managed plan.
OpenAI
GPT-4o · o-series
Anthropic
Claude Opus · Sonnet · Haiku
Google Gemini
2.0 Flash · 1.5 Pro
Groq
Llama · Mixtral
Mistral
Large · Codestral
// features
Everything it does, on top of any app
A focused toolkit — only what actually ships in the app today.
Streaming markdown & code
Answers stream in live with formatted markdown and syntax-highlighted code blocks — copy them straight out.
Screenshot & image analysis
Capture the screen with a shortcut and ask about it. Works with any vision-capable model you've configured.
Live transcription
Toggle on system-audio transcription and let what you hear feed the assistant in real time.
Knowledge files
Upload PDF, TXT, or MD and tag them — Resume, JD, Intro, Project — so the AI grounds answers in your own context.
Local conversation history
Every chat is saved on your disk. Search, rename, and delete past conversations; nothing leaves your machine.
Shape the voice
Tap communication-style pills or write a custom system prompt to set exactly how answers come back.
Plus window UX: resizes to fit, opacity slider, position lock, and per-OS shortcut labels.
// knowledge
Give your AI the context
Attach your own files, tag what each one is, and every answer is grounded in your material — not generic guesses.
- resume.pdfResume
- job-description.pdfJD
- project-notes.mdProject
01Attach
Drop in a PDF, text, or markdown file — or several at once.
02Tag
Label each one: Resume, JD, Intro, Project, or a tag of your own.
03Grounded context
The AI reads them and grounds every answer in your material.
Tags become section headings in the prompt, so the AI knows exactly what each file is.
Drawing on your Resume and the JDyou attached, I'd lead with your payments-reliability work — it maps directly to their staff-level scope.
// use cases
One overlay, plenty of moments
The same window, grounded in whatever you give it — your files, your screen, what you're hearing.
Interview prep
Upload your resume and the job description as knowledge files, then rehearse answers grounded in your own experience — in your words.
Meetings & calls
Turn on live transcription so the assistant follows along, and pull up context or summaries without leaving the call.
Coding help
Screenshot an error or paste a stack trace and get an explanation plus a fix, with code blocks you can copy.
Research & writing
Feed it source PDFs and notes as knowledge files, then ask, draft, and refine against material it can actually cite.
// shortcuts
Keyboard-first, by design
Global shortcuts work even when the overlay isn't focused.
A full list is also in the app — open the ⌨ Shortcuts panel in the header.
// privacy
Private because there's nothing in the middle
On bring-your-own-key your prompts never touch our servers. Either way, conversations stay on your disk and the overlay stays off your screen shares — honest about what that does, and doesn't, mean.
Your key, your provider
On Bring-Your-Own-Key, prompts go directly to the AI you chose with your own key — nothing routes through us. Managed plans pass through our proxy only to run the model and meter usage.
Local-first storage
Settings, conversations, and knowledge files live on your disk via electron-store. Delete the config file and the app resets to factory state.
Stays off your screen shares
The overlay opts out of screen capture and recording on macOS and Windows using Electron's content protection, so it won't show up when you share your screen.
What this is not: staying off screen shares is OS-level capture exclusion — not encryption, and not a guarantee against every recording, monitoring, or proctoring tool. Your chosen AI provider still receives your prompts under their own policies.
// pricing
Two ways to run it. Your call.
Bring your own provider key and keep prompts fully private — or let us provide the key and skip the setup entirely.
BYOK
Use your own API key.
- Connect your own OpenAI, Claude, Gemini, Groq or Mistral key
- Full overlay: streaming answers, code blocks, screenshots
- Live transcription & tagged knowledge files
- Conversation history stored on your disk
- Prompts go straight to your provider — never through us
You also pay your AI provider for usage.
Pro
We bring the key — no setup.
- No API key needed — model usage included
- Top models: GPT-4o, Claude Sonnet, Gemini
- Everything in BYOK
- Generous monthly usage
- Priority support
We handle the provider key for you.
Max
Everything, at full power.
- Everything in Pro
- Highest monthly usage limits
- Frontier models, incl. Claude Opus
- Earliest access to new models
- Priority support
We handle the provider key for you.
Billed monthly in INR · no free tier · cancel anytime
// faq
Straight answers
What does it cost?
Plans start at ₹599/month for Bring-Your-Own-Key — you connect your own provider key and pay that provider for usage. Managed plans (₹2,999 and ₹4,999/month) include the model usage, so you don't need a key. There's no free tier.
What's the difference between the plans?
BYOK (₹599) uses your own OpenAI / Claude / Gemini / Groq / Mistral key and keeps prompts fully private. Pro (₹2,999) and Max (₹4,999) include the key and usage — Max adds the highest limits and frontier models like Claude Opus.
Do I need my own API key?
Only on the BYOK plan, where you bring your own. On Pro and Max we provide the key — just sign in and start.
Where is my data stored?
On your machine. Settings, conversations, and uploaded knowledge files are saved locally via electron-store. On macOS that's ~/Library/Application Support/Gupt AI/; on Windows it's under %APPDATA%. Delete that file to reset the app.
Do you see my prompts?
On BYOK, no — your prompts go straight to your provider with your key and never touch our servers. On managed plans (Pro / Max), requests pass through our proxy so we can run the model and meter usage; we don't sell them or train on them.
Which providers and models can I use?
OpenAI, Anthropic (Claude), Google Gemini, Groq, or Mistral. Each has several models, and the in-app picker auto-saves when you switch. On BYOK you choose by adding that provider's key; on managed plans the top models are included.
Is it really invisible during screen sharing?
The overlay opts out of screen capture and recording on macOS and Windows via Electron's content protection, so it won't appear in a typical screen share. That's all it is — OS-level capture exclusion, not encryption and not a guarantee against every monitoring or proctoring tool.
Put a private AI on top of everything
Create an account, pick a plan, and summon it anywhere.
Plans from ₹599/mo · cancel anytime