Alala
End-to-end encrypted messenger with AI inside chats
7 followers
End-to-end encrypted messenger with AI inside chats
7 followers
Alala is an end-to-end encrypted messenger with a private AI inside your chats — it translates, summarizes long threads, drafts replies and searches by meaning. Zero-knowledge by design, plus encrypted calls, channels and live video. Web, iOS, Android.
How does the AI actually run inside an encrypted chat without seeing the message content? That part feels hard to square.
@nurantrnakcuhu
the AI does not "run inside" your encrypted chat and does not decrypt your messages. End-to-end encryption is never weakened or bypassed. The keys live only on your devices — the server only ever holds ciphertext and cannot see the plaintext.
The AI doesn't touch your conversation at all until you press the button (AI Assist) and explicitly consent for that specific chat. And even then, it isn't the server "peeking" — it's your client, the only holder of the keys, handing the AI a copy of the relevant messages for a single request. That copy doesn't travel in the clear either: your client encrypts it under a separate ephemeral key, it goes to an isolated processing zone, is decrypted only in memory for the moment the answer is generated, is never stored, never logged, and never used for training, and the result comes back encrypted only to you — the other person never sees it. The key is wiped from memory once processing is done.
So the boundary is simple: without your explicit action, the AI has no access to your conversation whatsoever. When you do invoke it, it's a deliberate hand-off of a copy — much like pasting text into a tool yourself — but with encryption preserved along the whole path and nothing retained.
And we'll be honest where others rely on slogans: classic E2EE is fundamentally incompatible with a server-side AI that needs plaintext to generate a reply. So we say it plainly — during processing the plaintext briefly exists in server memory (protected by TLS + at-rest encryption + isolation + strict governance). Our roadmap's next step is a confidential mode (TEE with remote attestation) that seals even that brief window in hardware. The separate "AI chat" assistant, by contrast, is not E2EE and we don't claim it to be — it's a conversation with the AI as its own participant.
For comparison: Signal has no AI in E2EE chats at all, and WhatsApp's "Meta AI" is a separate, non-encrypted chat where your message goes to Meta. With us, AI inside protected conversations runs only on explicit consent, ephemerally, in an isolated zone, and without training on your data.
Launched live streaming across web and mobile with previews, HD video, comments, pinned messages, and notifications.
Improved video notes and media-message playback.
Added secure call transcription across iOS, Android, and Web.
Added per-chat screenshot protection controls.
Improved encrypted messaging reliability and multi-device synchronization.
Enhanced device linking and message-history recovery.
Added phone verification, phone-number contact discovery, and native share support.
Upgraded the business workspace with AI summaries, dashboards, integrations, and legal-update alerts.
Improved AI legal answers with current sources, effective dates, and verifiable references.
Updated the mobile wallet with PIN access, biometrics, account recovery, and transaction history.
Refined the interface, themes, search, Arabic localization, and accessibility.
Updated public documentation and completed scheduled stability and security improvements.