Your anonymity — the full story

Step by step: the journey of your confession

1. You open this site. If you came from an ad, the link may carry utm_source, utm_medium and utm_campaign. Only pre-approved (allowlisted) values are read into your browser's memory. Everything else — click IDs like fbclid and gclid — is discarded at once and never reaches our server. The address bar is cleaned immediately.
2. You read our promise and agree. That records a single 'yes' in your browser's memory — no cookie, no storage.
3. You speak, write, or upload a file. Everything stays in your browser's memory — refresh or close the page and it is gone. When you upload a document, our server extracts the text within that same request and shows it to you; the file and its metadata (author, company, dates) are thrown away at once. Nothing is saved. When you choose an audio file, your browser turns it into plain sound on your device first — the file name and every detail stored inside the file (tags, dates, the device it came from) are left behind — and that sound is then treated exactly like a recording.
4. On Review you see exactly what will be sent. If anything looks like a phone number, email, Aadhaar, vehicle number or PAN, you can remove it with one tap.
5. When you tap 'Send to the universe', only these things reach us: the confession (audio or text), how you said it, the language, your consent and the policy version, a campaign name (if there was one), and the time — rounded to the hour. Our server does not read, log, or store your IP address, browser details, or request headers.
6. Audio is stored in private Vercel Blob storage under a random UUID — tied only to that confession, never to you. It has no public link, and none can be made. Only our team — after signing in with their email and password — can listen to it, streamed through our own server. The one other place it goes is the transcription step described next, which keeps nothing.
7. After you send, the confession text — or, for a voice confession, the recording — is sent through OpenRouter to an AI model running on Amazon Bedrock or Google Cloud. Only endpoints that keep nothing (zero data retention) and never train on it are used. For a voice confession this makes the transcript. The model then writes a short summary of the confession for our team, saved with it so the team can read each one at a glance; it is not shown to you. Nothing about you goes with it — no IP address, no browser details, no request headers.
8. The moment it is sent, the draft is erased from your device — no draft, no ID, no cookie. Going back does not show the confession again.
9. Our small team listens or reads, removes anything that could identify a person — a name becomes a relation, a city becomes a region, office and school names are dropped — and decides whether it will be shared. Every action the team takes is recorded. Nothing about you is recorded, because we hold nothing about you.

What we keep

• Your confession — audio, text, or the text taken from a file
• A short AI summary of your confession, written for our team
• The time it was sent, rounded to the hour
• How you said it — voice, text, or file — and the language
• Your consent, and the policy version
• A campaign name if you came from an ad — allowlisted names only
• Editorial notes added by our team — themes, language, decision

What we never keep

• Name, phone number, email, or any contact detail
• IP address, device ID, browser fingerprint, user agent
• Cookies, storage identifiers, or any session after you send
• Ad click IDs — fbclid, gclid and the like — or any attribution tied to you
• Uploaded files, or their metadata (author, company, dates, device)
• Location or timezone — precise or approximate

Where things are kept

This site runs on Vercel; our server code runs in the Mumbai region.

The confession text and the little kept alongside it live in a Neon Postgres database hosted in Singapore — that is, outside India. We say so plainly. Nothing stored there can identify you.

Audio is kept in private Vercel Blob storage. It has no public link.

The transcript and the summary are made by AI models reached through OpenRouter, running on Amazon Bedrock or Google Cloud. Every request allows only zero-data-retention endpoints on those two hosts and refuses data collection, so the services that run those models keep nothing. The summary is stored with the confession, in the same Neon database.

The truth about server logs

Our hosting provider (Vercel) keeps some standard server logs, as every website's host does — for a short time, to run its own systems. Our domain is routed through Cloudflare, which carries each request to Vercel and keeps its own brief logs under its own terms. Neither is in our hands, so we do not hide it.

We do not look at these logs, we do not send them anywhere (no log drains), and they cannot be joined to a confession — because nothing is stored with a confession that could be matched against a log. Even the time is rounded to the hour.

Our own application never writes request data to its logs. When something goes wrong, only a short error code is recorded.

Bots and spam

There is no bot check on this site: we tried one, and it turned real people away. Any spam protection we add will run at the edge — Vercel or Cloudflare — outside our application, which never sees your IP address and stores nothing for rate limiting.

If that ever changes — a rate limit, or a separately disclosed check such as Cloudflare Turnstile — this page will say so.

Your voice

Your voice is never published. A voice is an identity.

If your confession is shared, it appears as text or in someone else's voice — the host's voice-over or a computer voice. A transcript is made for editing: the recording is sent through OpenRouter to an AI model on Amazon Bedrock or Google Cloud, only on endpoints that keep nothing (zero data retention) and never train on it.

The original recording is permanently deleted 30 days after a decision is made on the confession.

Why we cannot identify you

Because we hold nothing about you. No name, no number, no IP, no device — nothing.

If someone asks us who sent a confession, our honest answer is: we do not know, and we cannot find out.

That is also why we cannot act on a 'delete my confession' request without the withdrawal code — we have no way to check that it is yours.

How long things are kept

• The draft on your device: only while the page is open. Sending or closing the page erases it at once.
• Uploaded files: never stored. From a document only the text is kept; from an audio file only the sound, rebuilt on your device without the file or its details, and then kept like a recording.
• Original audio (Vercel Blob, private): permanently deleted 30 days after a decision (publish, reject, or withdraw).
• Confession text: kept while it is in review or published. Deleted when withdrawn with the code (soon).
• The short AI summary: kept with the confession text, for as long as the text is kept.
• What is sent to the AI model through OpenRouter — the text or the recording: not kept. Only zero-data-retention endpoints are used.
• Database restore history (Neon): encrypted, at most 30 days. Anything deleted leaves it within that window. No separate backup copy of Blob storage is kept.
• Vercel's and Cloudflare's standard server logs: kept briefly, under their own terms. We never forward them.

Taking a confession back — coming soon

After you send, you will see an 8-character code exactly once. We keep only a hash of it — tied to that confession, never to you.

Enter the code: if the confession has not been published yet, it is deleted immediately. If it has been published, it goes to takedown review within 48 hours.

If the code is lost, we cannot help — we have no way to check that a confession is yours. Keep the code somewhere safe.

This feature is being built. Until then, look carefully at Review before sending — there is no other way to take a confession back.