Clearbase

Why Clearbase

The internet has a business model problem

Most "free" software is free because you pay with your attention, habits, and identity. We built Clearbase on a different model: you pay a fair price, we build tools that serve you.

Honest about what we send and where

Your audio and full transcripts never leave your device. Only a hashed version of short question text reaches our servers for AI processing. We cannot read it, no one else can, and we do not sell it. We monitor token usage only to detect abuse. You can delete your account and all data at any time.

No ad networks, ever

A privacy company that uses Google AdSense is not a privacy company. We will never embed ad networks that track your behavior across the internet.

Transparent by default

We publish exactly what API calls are made and what data is sent. No vague policy language. If it touches a server, we tell you which one and why. You will find the specifics in each product's dedicated section.

One team, no investors

A small but dedicated team of engineers with no VC pressure to monetize users. No board pushing for growth at your expense. Your interests and our interests are exactly the same.

Delete anytime, for real

Request account deletion and every record is purged within 30 days. No "we'll archive it for 90 days" asterisks.

No surveillance capitalism

We earn revenue from subscriptions and privacy-respecting contextual ads on the free tier. No behavioral tracking. No data sales. No ad networks that profile you. What you pay for is what you get.

How it works

Designed to protect you at every step

Here is exactly what happens when you use Whispr.

You install the desktop app

Whispr runs natively on Windows. It uses OS-level rendering so it's invisible to screen-capture software. Your interviewer sees only your screen, not Whispr.

Audio captured locally stays on your device

Whispr listens to your meeting audio directly on your machine. The audio itself is never uploaded or transmitted anywhere. It is processed and discarded locally.

Question text is sent for AI processing

When a question is detected or typed, the question text is sent to a third-party AI provider (Groq or Gemini) and also briefly logged on our server for abuse detection and service quality monitoring. Short transcript snippets are sent for auto-detection (up to 2,000 characters) and for generating answers (up to 3,000 characters). No audio. No full transcripts. The answer appears in your overlay within seconds.

Session ends, audio and transcripts discarded

When your session closes, audio and transcripts are permanently discarded from your device. The only information we store is your account, basic usage records (for abuse monitoring and server traffic costs), and any conversations you choose to save, connected directly to your account for your eyes only. All of this is controlled by you. Delete your account and everything tied to it is purged within 30 days, no middleman, no exceptions.

Our product

Whispr: AI interview and meeting assistant, invisible to screen share

Real-time AI guidance for interview preparation and live professional conversations. Only you can see it. Everyone else on the call sees your screen as normal.

Completely invisible to Zoom, Teams, and Google Meet screen share
Audio stays on your device. Never uploaded or transmitted anywhere
Real-time answers with adjustable transparency overlay
Built for technical interviews, sales calls, and any live meeting
Free tier available. No credit card required to start

Join the waitlist, be first to know

Which servers handle your data: Our own server for account authentication and session metadata. Groq for free-tier AI responses. Gemini (paid API) for Pro-tier responses.

Whispr · Active session

💬 "Walk me through how you'd design a rate limiter."

A rate limiter can be implemented with a token bucket or sliding window approach. I'd store request counts in Redis with a TTL equal to your window size, increment on each request, and reject once over the limit. For distributed systems, use a Lua script to make the increment-and-check atomic...

Screen share safe · Audio stays on device

Privacy promise

Exactly what we do with your data

No vague policy language. Concrete commitments, in plain English.

Audio & video

Meeting audio never leaves your device. Captured locally, processed locally, discarded when the session ends.

AI processing

Your audio stays on your device. Short text snippets are sent to AI APIs for question detection and answering. Never audio. Never full transcripts. Question text is briefly logged on our server to detect abuse and monitor service quality, this is disclosed, not hidden. We have configured both providers for minimal retention.

Groq (free tier users): Inference requests are not retained by default. Temporary logging occurs only when troubleshooting errors or investigating abuse, retained for up to 30 days. Groq data policy

Gemini (Pro tier users): Pro tier requests are routed through Google's paid API. Google does not use your prompts to improve its products. Prompts are logged for up to 55 days solely for abuse monitoring and are not used to train AI models. Gemini data policy

Usage monitoring

We run an internal admin dashboard that tracks session counts, LLM call frequency, and token usage per account. We are transparent that this exists. Its purpose is to detect abuse such as API abuse, rate limit violations, and spam signups, so that the free tier stays genuinely free for real users. This data is never sold and is not shared with third parties. You can request a full export or deletion of your usage data at any time.

Analytics

We use Plausible Analytics. Privacy-respecting, no cookies, no fingerprinting. No Google Analytics, no Facebook Pixel.

Advertising

Free-tier ads use EthicalAds. Contextual targeting only, no tracking cookies, no behavioral profiling. We will never use Google AdSense.

Data sharing

We don't sell your data. We don't share it with data brokers.

Account deletion

Request deletion and every record we hold about you is purged within 30 days. No exceptions, no backup archives retained.

Open by design

Don't trust us, verify us

Any privacy company can write a policy. We show you the actual code. Here's exactly what Whispr sends to AI providers and how it stays invisible on your screen.

llm_service.py What we send to AI

# From llm_service.py — what we send to Groq / Gemini
# Audio? Never touches this function.
# Full transcript? Truncated to last 3,000 chars max.
# Your identity? Not included.

async def generate_answer(question, doc_context, transcript, ...):
    parts = []

    # Prepared Q&A pairs if you loaded a document
    if prepared_qas:
        parts.append(f"## PREPARED Q&A\n{prepared_qas}")

    # Uploaded document or prior conversation context
    if doc_context:
        parts.append(f"## DOCUMENTS\n{doc_context}")

    # Last 3,000 chars of transcript only — not the full session
    if transcript:
        parts.append(f"## TRANSCRIPT\n{transcript[-3000:]}")

    # Your question — typed or auto-detected
    parts.append(f"## QUESTION\n{question}")

    # That is everything sent. Nothing else leaves your machine.
    return await _call_llm(provider, model, system, "\n\n".join(parts))

stealth.rs How we hide from screen capture

// From stealth.rs — how we hide from screen capture
// Windows API: WDA_EXCLUDEFROMCAPTURE (0x00000011)
// This OS-level flag makes Whispr invisible to
// screen recording software. Not a trick. Not a filter.
// The OS itself excludes this window from capture.

extern "system" {
    fn SetWindowDisplayAffinity(
        hwnd: isize,
        affinity: u32,
    ) -> i32;
}

pub fn enable_stealth(hwnd: isize) -> Result<(), String> {
    let result = unsafe {
        SetWindowDisplayAffinity(hwnd, 0x00000011)
    };
    // Requires Windows 10 version 2004 or later
    if result == 0 { Err("Requires Windows 10 2004+".into()) }
    else { Ok(()) }
}

These are real excerpts from Whispr's codebase. We show this because the strongest privacy claim is one you can verify yourself. If you have questions about any specific data flow, reach out directly and we will walk you through it.

What's coming

More privacy tools on the roadmap

We build only when existing usage data proves demand. Every product below follows the same privacy-first model as Whispr.

Local Password Vault

In production

A password manager that stores everything encrypted on your device, no company holding your keys, ever. Available purely offline, with an optional connected tier for sync.

Meeting Summarizer

Exploring

Automatically generate clean summaries of your meetings after each call. Audio processed locally, only the transcript text used for summarization. Never your full recording.

Get in touch

Join the waitlist

Be the first to know when Whispr launches. Early waitlist members get priority access and a launch discount.

Have feedback, a suggestion, or something we should know? Use the Contact tab to reach us directly.

Your details go directly to our inbox. No third-party form services, no marketing automation, no drip campaigns.

New Jersey, USA

First name

Last name

Email address

What are you hoping to use Whispr for? (select all that apply)

Interview preparation

Practice answering technical and behavioral questions with real-time AI guidance before your next interview.

Live meeting assistance

Get on-the-spot AI support during active calls, presentations, or client conversations.

Sales calls

Stay sharp during client pitches and negotiations with discreet real-time prompting.

Privacy and data security

Interested in Clearbase specifically because of how we handle your data, not just the AI features.

Just exploring

Not sure yet. You want to learn more about what Whispr does before committing to anything.

✓ You're on the list! We'll email you the moment Whispr launches.