Viper Messenger
A private messenger built on TON. Messages go device to device whenever they can, everything is end-to-end encrypted, and your account is a key — no phone number, no email.
On-device AI: gpt-oss-20b runs locally on your Mac (Apple silicon, 24 GB) — answers from Viper's guides, nothing leaves the device. Smaller open models on iPhone and iPad.
iPhone · iPad · Mac
Private by how it is built
Not a policy promise: the design leaves little for a server to know.
Direct, device to device
Messages and files go straight between devices whenever possible, with no server in the middle. When a direct link is not possible, the relay passes on what it cannot read.
End-to-end encrypted
Only you and your peer can read your messages — over a direct link, over Bluetooth and through the relay alike.
History lives on your device
Chats are stored on your device, encrypted. The relay deletes each message once it is delivered and keeps no history.
No phone number or email
Your crypto key is your account. Start with a 24-word recovery phrase, no sign-up; the same phrase brings your account to a new device.
Voice, video circles, large files
Voice notes, video circles up to 30 seconds and large files. A transfer picks up where it stopped if the connection drops.
.ton usernames
Use a TON DNS domain you own on-chain as your name. Type a name with .ton in Search to find who holds it.
TON wallet built in
Your recovery phrase also makes a non-custodial TON wallet: one set of words restores both your account and your wallet.
Viper Pro
Anon Mode takes you out of Search while your chats keep working, and Hide Online Status shows you offline to everyone.
People nearby, no internet
Find Viper users around you over Bluetooth and chat with no network at all. See TON Drop.
On-device AI
An AI chat that runs on your device and knows Viper. It drafts messages, events and notes — nothing is sent until you confirm. More.
Also in the app: audio and video calls on iPhone (end-to-end encrypted, through the relay), Encrypted Notes, Face ID lock and an encrypted chat backup.
A look inside






Viper 1.5.34 views rendered in the iOS simulator with sample contacts and messages. The AI draft screen needs gpt-oss-20b, which runs on a Mac.
Bluetooth to meet. TON to stay connected.
Two Viper users next to each other find each other and start talking straight away — no internet, no phone number, no username. Here is exactly where it stands.
- Works now
Discovery over Bluetooth. People Nearby shows the Viper users around you and roughly how far away they are. Everyone on the list has proven they own their account.
- Works now
Fully offline over Bluetooth. Messages, voice notes, video circles and files go phone to phone with no network and no server.
- Works now
Bluetooth, then direct over the internet. Phones can exchange IP addresses over Bluetooth and then connect directly, P2P over the internet with TON — no server in between.
- In progress
A stable link out of range. When either phone's IP address changes, the direct link breaks today. Keeping it alive needs TON DHT to find each other again; that is the work in progress for TON Drop to fully work as described.

A stock open model. Viper's harness. Your device.
On a Mac, Viper runs gpt-oss-20b locally — no cloud, no AI account, your questions never leave the device. It answers from Viper's own guides and shows the sources under every answer, so you can check it in one tap.
Answers translated from the original Russian session.
- No training. The model's weights are exactly as released. Viper changes what it sees, not what it is — swap in a better model and the knowledge stays.
- Sources under answers. Answers come from Viper's in-app guides, and the guides they used are listed right below.
- Drafts, not actions. Ask it to message someone, create an event or save a note: it prepares a draft, and nothing happens until you confirm.
- You choose what it knows. Turn Viper's built-in knowledge on or off, add your own notes.
Not always right yet: the strict test counts an answer correct only when every Viper detail in it is true. A new harness reaches 61% on the test bench and is not in the app yet. gpt-oss-20b needs a Mac with Apple silicon and 24 GB of memory; iPhone and iPad run smaller open models. The full write-up →
