Validate before committing
Let’s run a quick end-to-end test in the actual client environment right now, not a guess. Then we either commit Friday with confidence, or we give them a realistic date.
AskSharper transcribes your call on your own machine with Whisper. Press F2 and it suggests your next line using your own Anthropic key; your audio never leaves your PC.
Windows 10 & 11 only · Requires your own Anthropic API key · Free during beta · Audio stays on your device
“It worked fine when we ran it last week.”
“I don’t want to promise a date we can’t hold.”
Let’s run a quick end-to-end test in the actual client environment right now, not a guess. Then we either commit Friday with confidence, or we give them a realistic date.
If we tell them Friday and it breaks, who owns that failure? Before we commit, we need agreement on acceptable risk tolerance and a rollback plan.
When you say ‘not finished yet,’ what’s actually blocking completion? Is it code, testing, integration, or something else?
AskSharper is a local AI meeting assistant for Windows. It listens to your mic and system audio and transcribes on your own device with Whisper. When you press F2, only the recent text is sent to Claude using your own Anthropic key to suggest what to say next. Audio never leaves your PC.
One session, start to finish.
Every screen below is the shipping app on Windows 11 — nothing staged or redrawn.
AskSharper installs as a tray app. No window to manage, no bot to invite to the call, nothing for other participants to see.
AskSharper is running
It lives in your system tray, the dot icon near the clock (bottom-right). Right-click it to start a session. If you don’t see it, click the ^ arrow to show hidden icons.
Click to dismiss.
Pick a meeting type and, optionally, a topic. That’s what shapes the kind of suggestion you get back.
Start a new session
Your assistant will listen alongside you and suggest what you could say when you press F2.
Mid-call, press F2. A few options appear over your screen — a sharper question, a risk worth naming, a word worth defining. Nothing is spoken or sent for you.
“When we tell the client Friday, are we saying feature-complete, tested, or deployed? That changes what we’re actually promising.”
— one of four suggestions from the session above.
When we tell the client Friday, are we saying feature-complete, tested, or deployed? That changes what we’re actually promising and what happens if we slip.
Before we commit to Friday, we need to be explicit: shipping untested code creates technical debt and potential production issues. What’s the actual test coverage gap?
Stop the session and AskSharper writes a summary and follow-up items — the AI meeting notes you’d expect, produced and stored on your own machine.
The team discussed whether to commit to a Friday deadline for client delivery despite incomplete testing. There was uncertainty about readiness, but the decision was made to tell the client Friday is feasible and work to meet that timeline.
Windows 10 & 11 only · You’ll paste your own Anthropic API key on first run.
Note-takers record. AskSharper tells you what to say.
Passive tools transcribe the call and hand you a summary afterward. AskSharper works in the moment: press F2 mid-conversation and it reads the recent transcript and suggests a sharper question, a risk to surface, or a point you might be missing, while it can still change the outcome. That real-time coaching is the one thing no note-taker does.
Download, run, paste your key, press F2.
Download the Windows installer.
Run the installer (Windows 10 or 11).
Paste your Anthropic API key when it asks.
Press F2 in your next meeting and see what to say next.
Windows 10/11 installer. Free during beta. No account needed.
Three steps. Audio stays on your device throughout.
Start it next to Teams, Zoom, Meet, or Discord. It uses OS-level audio, so there’s no bot to invite.
Speech-to-text runs on your own CPU and turns audio into text right on your device. Nothing is uploaded.
Only recent text (never audio) goes to Claude on your own key. You get options for what to say next.
Works with
Now in 28 languages, and hidden from your screen share.
Pick your language in Settings, or in the row under Topic when you start a session. One switch sets both the language it transcribes and the language it answers in, so it can never listen in Spanish and suggest in English.
English is the default. The rest include Spanish, French, German, Italian, Portuguese, Dutch, Polish, Ukrainian, Russian, Turkish, Arabic, Hebrew, Hindi, Chinese, Japanese, Korean, Vietnamese, Thai, and Indonesian.
It is a setting, not another download: the Whisper model already in the app is the multilingual one. The 28 are the ones it handles well, not every language Whisper will attempt.
The floating pill and the suggestion panel sit on top of everything, which includes your own screen share. So AskSharper asks Windows to leave its windows out of any capture: whoever is watching sees what is behind them, not a black rectangle.
It is on by default, because the opt-in version means everyone broadcasts their suggestions exactly once. Switch it off in Settings, under Privacy.
Needs Windows 10 version 2004 or later; on older builds the windows stay visible in captures. It does not hide your screen from a camera pointed at it.
AskSharper vs cloud note-takers and Cluely
| Capability | AskSharper | Cloud note-takers (Otter, Fireflies) | Cluely-style real-time tools |
|---|---|---|---|
| Where audio is processed | On your device (local Whisper) | In the cloud | Typically in the cloud |
| Pricing model | Bring your own key, no per-seat fee | Per-seat SaaS subscription | Per-seat SaaS subscription |
| Works with any meeting app | Yes: OS-level audio, no bot | Often via a per-platform bot or plugin | Varies by tool |
| Tells you what to say in real time | Yes: press F2 | No: records and summarizes | Yes (their category) |
| Per-session consent reminder | Yes, built in | Varies by tool | Varies by tool |
Comparison reflects the general categories, not every product; features of other tools change over time.
Your audio never leaves your device.
The only thing that ever leaves your PC is the recent transcript text, and only when you press F2, sent straight to Anthropic on your own API key. Your audio never does: it’s captured and transcribed entirely on your machine.
Speech-to-text runs on your CPU. AskSharper has no backend that touches your audio. Ever.
Text goes directly from your device to Anthropic, on your account. AskSharper doesn’t proxy, store, or see that traffic.
A consent reminder is built in. The rest is your call.
Recording and transcription laws vary by state and country. AskSharper shows a consent reminder at the start of every session as a built-in feature, but you are responsible for actually notifying every participant, especially in all-party-consent jurisdictions like California, Florida, Illinois, Massachusetts, and Washington.
For anyone who has to think and talk at the same time.
Discovery, scope, kickoffs, client check-ins. Lead the conversation and still land a sharper follow-up question.
On either side of the table, surface a risk you’d otherwise miss before the moment passes.
Keep momentum and don’t let a constraint slip past while you’re busy carrying the call.
Built by an independent dev who kept freezing in client calls, and wanted a second brain that stayed on his own machine.
Frequently asked questions
No. AskSharper transcribes your mic and system audio locally on your own machine with Whisper; the audio is never uploaded. Only when you press F2 does the recent text (never audio) go to Claude, sent directly from your device to Anthropic on your own API key.
Yes. AskSharper uses your own Anthropic API key, so suggestions go directly from your device to Anthropic on your account. AskSharper itself is free during beta; you pay Anthropic per use on your own key, and there’s no per-seat subscription to AskSharper. You can create a key at console.anthropic.com.
Note-takers record and summarize after the fact. AskSharper does that locally too (you get a transcript and an end-of-session summary), but its point is real time: press F2 and it tells you what to say next, in the moment. No passive note-taker does that.
Yes. AskSharper captures audio at the operating-system level, so it works with any meeting app (Zoom, Microsoft Teams, Google Meet, Discord, Webex) with no bot to invite and no per-app plugin to install.
Yes. AskSharper listens and answers in any of 28 languages, including Spanish, French, German, Portuguese, Japanese, Korean, Chinese, Hindi, Arabic, Ukrainian, and Russian. Pick one in Settings, or in the row under Topic when you start a session. The same choice sets both the transcription language and the language of the suggestions, so it can never listen in one language and answer in another. English is the default. The list is the set the shipped Whisper model handles well, and the language is read once when the session starts.
Not with the default setting. AskSharper asks Windows to exclude its own windows from screen capture, so the floating pill and the suggestion panel do not appear in what you share or record; the other side sees whatever is behind them, not a black rectangle. It is on by default, and you can switch it off in Settings under Privacy. It needs Windows 10 version 2004 or later; on older builds the windows stay visible in captures. It cannot hide your screen from a camera pointed at it.
It depends on where you and the other participants are. Some places require all parties to consent (for example California, Florida, Illinois, Massachusetts, and Washington). AskSharper shows a consent reminder at the start of every session, but you are responsible for actually notifying everyone.
AskSharper is free during beta. You bring your own Anthropic API key and pay Anthropic directly for the API usage; there is no AskSharper subscription. Windows 10 and 11 are supported today.
F2 is the trigger. While you’re in a call, press F2 and AskSharper sends only the recent transcript text to Claude on your own key and shows a few suggestions for what you could say next: a sharper question, a risk to surface, a point you might be missing.
Not yet. AskSharper runs on Windows 10 and 11 today. A Mac version is planned for later. The fastest way to hear when it lands is the ‘Get notified’ link.
Install the local, private meeting assistant and press F2 in your next call.
Free during beta · Windows 10 & 11 · Mac later · Bring your own Anthropic key.
Know someone who talks for a living? Send them this: it’s free, private, and runs entirely on their own machine.
Updated 2026-07-17