AI has learned to sound exactly like us, and it's already in your DMs, your inbox, and your dating apps. Deckrd watches your conversations locally on your device and tells you when the other side isn't human. Know who you're really talking to.
Generative AI has crossed the line at which most people can no longer reliably tell. There are only two ways to solve the bot problem at scale. The platforms and governments are converging on one of them. We're building the other.
Iris scans. Personhood proofs. Government-linked credentials tied to every account.
Solves the bot problem at the cost of forcing every human onto a permanent identity rail. Once that rail exists, it doesn't come down.
Detection runs on the AI side of the conversation. Humans are never forced to prove themselves, but you can verify your voice to the people you trust.
Solves the bot problem without trading away human privacy to do it.
AI is getting better at sounding human, in your inbox, your DMs, your dating apps, even your family's phone. Deckrd is the layer of truth between you and the bots.
Every video call app now has an uninvited AI note-taker. Some of them are training voice models on you. Deckrd Protect stops your voice being cloned or transcribed, and Verify lets family hear it's really you.
The perfect match who always says the right thing. The new follower sliding into your DMs. When the chemistry feels too good to be scripted, Deckrd lets you know if it actually is.
When your parents get a scam call with a cloned voice of their child, or a text "from the bank" asking them to verify their account, Deckrd Detect catches the bots, and Verify shows trusted family “✓ Verified” when it's really you.
The candidate acing your interview in real time, is it them, or their AI? Make sure the person you're hiring is the one actually doing the talking.
Humans are never forced to prove themselves. Verify is opt-in cryptography for your trusted circle, so family and close contacts can hear it's really you, speaking live, not a recording or a clone.
While you talk, your Mac stamps your voice with a continuously refreshed signature only your device can produce. Contacts you've paired with see ✓ Verified. A perfect AI clone still can't forge it, because the clone doesn't have your key. No voiceprint is ever stored.
Verify is a private, voluntary trust layer for people you choose, the opposite of forcing every human onto a permanent identity rail. Pair once out-of-band, then every call is checked automatically.
Pair a trusted contact: exchange identity keys and confirm a short SAS code out-of-band, like Signal safety numbers.
Start a call: Deckrd handshakes over a secure relay when you connect, and identity is confirmed within seconds of first speech.
See the verdict: “✓ Verified” for the real person speaking live, “⚠ Unverified” for a recording, relay, or clone.
Add Deckrd to your browser and every conversation you open from then on gets checked, each one given a verdict with a confidence score.
Grab the browser extension from the Chrome Web Store or Firefox Add-ons. One click, no account, no setup. Launch the extension and Deckrd goes to work.
Our ensemble engine evaluates every conversation locally on your device. Nothing leaves your machine.
A clear, instant signal the moment the engine crosses its confidence threshold. You'll see Human, AI Detected, or Analysing, so you're never left guessing.
Deckrd installs once and runs quietly in the background. When you open a conversation, it surfaces a verdict the moment one is reachable, without you ever asking, without your messages ever leaving your device.
Other tools require you to copy and paste a suspicious message into a checker. By then it's too late. Deckrd watches in real time and tells you. You never have to suspect: we tell you when there's something to suspect.
Eleven independent local detectors running simultaneously. When generators get better at one signal, the others still hold.
Patterns analysed on your device. Nothing leaves your machine. Deckrd sees structural shape, never message content.
Pasting messages into a third-party checker means the user has to already suspect a bot to find one, and means the message is uploaded somewhere. Deckrd watches the conversation as it happens, on-device, and tells you when it sees something. That's the category split.
It will arrive as a conversation: helpdesk impersonation that adapts in real time, synthetic candidates in your hiring pipeline, cloned executive voices on the phone. Your stack inspects links, attachments, and content. Nothing in it asks whether the other party is human. Deckrd answers that question in real time.
Deckrd flags synthetic counterparties in the chat surfaces your people already use, on-device, without reading message content.
The Detect engine ships as an SDK and REST API for text channels. Voice detection is developed and pending release.
Deckrd Verify gives executives a cryptographic liveness signature that a perfect clone can't forge, because the clone doesn't have their key.
We're onboarding a small number of design partners ahead of general availability, with direct input into the enterprise roadmap and preferred pricing at launch. Talk to us: [email protected]
Deckrd is three capabilities on one engine: Detect, Protect, and Verify, scaling across chat, voice, video, and every device where conversations happen.
Free during the open beta. Installing takes seconds, with no account and no friction.
Detects AI across the chats you already use. ChatGPT, Discord, WhatsApp Web, Messenger, LinkedIn, X DMs, and more.
Detect AI on calls and in native chat apps. Protect your voice on Zoom, Meet, Teams, and every app using your mic. Verify a voice is genuine to family and trusted contacts, available now in the macOS alpha.
The same Detect engine that powers the extension, exposed as @deckrd/sdk. Protect and Verify APIs ship later. Free tier: 10,000 detections/month.
The browser extension covers desktop today. iOS and Android are next. Drop your email and we'll tell you the moment mobile and other capabilities ship. No spam, just product updates.