Load the context once
Pick a mode and fill only the fields that mode needs — JD and resume for interviews, scope and open decisions for client work.
- 3 modes, separate forms
- Length + tone per session
- Key encrypted via safeStorage
macOS overlay · BYOK
The answer is on screen before the pause gets awkward.
Kalfi listens to your call from your own Mac, catches the question, and writes one line you can actually say out loud. Point it at your own OpenAI or OpenRouter key and the app costs nothing to run.
- Modes
- Interview · Client · Chat
- Audio in
- System, mic, or both
- Your key
- Stays on device
Live transcript
system + mic
Suggested reply
gpt-4o-mini
Press Play, or hit Ask to force an answer on the last line.
Length
Tone
The loop
Context in, one speakable line out, summary at the end.
Not a transcript dump. Kalfi keeps the job or project in a thread, so the second meeting already knows what you promised in the first.
Listen the way the call runs
Mix meeting audio and your own mic, so a question you repeat out loud still gets an answer. Ask and Mic Ask override detection when it misses.
- System + mic mixing
- Force-answer any line
- Floats over fullscreen Meet
Close the meeting properly
Generate a summary, browse past meetings without hijacking the live one, then start the next meeting inside the same project thread.
- Per-meeting summaries
- Browse-only history
- Pagi / Siang / Sore / Malam labels
Stealth
Flip the switch and watch it leave the shared screen.
Kalfi uses a panel window with content protection and no meeting bot. Toggle stealth off below to see exactly what you would be leaking without it.
What the room sees — shared screen
clean
HK
You (presenting)
RNRina
DVDavid
AMAmir
What you see — your display
overlay on top
HK
You (presenting)
RNRina
DVDavid
AMAmir
Honest caveat: screen-capture protection depends on the OS and the conferencing app. Test your exact setup before you rely on it, and follow the rules of the room you are in.
Bring your own key
Do the maths before you sign a $150 subscription.
With BYOK you pay the model, not a middleman. Move the sliders to see what your month actually costs against the plans everyone else sells.
12
40
Assumes ~0.8 answered questions per minute, ~900 input and ~220 output tokens per answer. Token prices are public list rates and change often — treat this as an estimate, then check your provider dashboard.
Pricing
Two doors. Neither one is a trial that expires on you.
Technical users bring a key and stay free. Everyone else pays a flat fee and never sees a provider dashboard.
Free
BYOK$0 / forever
The full desktop app. You pay your provider directly for what you use.
- Interview, client meeting and chat modes
- System + mic capture, Ask and Mic Ask
- Meeting summaries and project threads
- OpenAI, OpenRouter or any compatible base URL
- Key encrypted on device, never sent to us
Pro
Hosted$12 / month
No key, no provider account. We carry the model bill and the setup.
- Everything in Free
- Hosted chat and transcription through our API
- Login on any Mac you own
- Cancel or change card in Stripe portal
- Monthly fair-use quota, shown in app
Landscape
What the other tabs in your browser are charging.
Public marketing and list prices, September 2026. Not a lab benchmark.
| Final Round | ParakeetAI | Cluely | LazyJobSeeker | Kalfi | |
|---|---|---|---|---|---|
| Live in-call answers | Pro plan only | Subscription or credits | Assist shortcut | After free session | Free with your key |
| Bring your own key | not offered | not offered | not offered | not offered | Primary path |
| Client / sales meetings | interview-first | Yes | Meeting-first | Yes | Dedicated mode + threads |
| Answer style control | preset | Model choice | preset | preset | Length + tone per session |
| Entry price | ~$25 / mo | ~$149.90 / mo | Paid plans | Paid after trial | $0 or $12 / mo |
Questions
The things people ask before downloading.
Does my API key ever touch your server?
No. On the Free plan the key lives in macOS encrypted storage and talks straight to OpenAI or OpenRouter from your machine. Our API only appears when you choose Pro, and then the provider key is ours, not yours.
What actually runs where?
The app runs on your Mac — audio capture, transcript, session history, all local. This page is static hosting. The Pro backend runs on a small VPS and only handles login, subscription state and the model proxy.
Is this allowed in my interview or meeting?
That depends on the room. Kalfi is built for calls where assistance is permitted — client work, internal meetings, sales, practice, language support. If an employer, school or client says no, respect it.
Which providers and models work?
Anything OpenAI-compatible: OpenAI directly, OpenRouter for model choice, or a custom base URL for your own gateway. Chat and transcription models are configurable per install.
Windows and Linux?
The build targets exist, but macOS Apple Silicon is what ships tested today. The overlay and stealth behaviour are tuned for macOS first.
Install it, paste a key, join your next call.
Apple Silicon DMG from GitHub Releases. The app is unsigned for now, so open it from the right-click menu the first time.