Update, 19 October 2026: ordinary direct messages are no longer end-to-end encrypted. Tied to one device, they could not follow people to a new phone, and in testing whole conversations became unreadable. They are now private to the two participants and encrypted in transit and at rest; everything below describes Secret chat, the opt-in conversation that is still end-to-end encrypted and stays on the phone or browser it was used on.
Conversations between a manager and an employee about expenses can touch salaries, health claims, travel plans, and disputes. That’s why Secret chats in Reimbilly are end-to-end encrypted: they are encrypted on your phone or in your browser before they leave it, and our servers store only ciphertext — a subpoenaed database dump holds nothing readable. One limit is spelled out below: the other person’s key comes from our server until you compare security codes with them. Channel messages are a different thing and this post says so plainly further down.
The building blocks
We use NaCl “box” encryption (Curve25519, XSalsa20 and Poly1305, through the TweetNaCl library). Every user gets a keypair generated on their own phone or in their own browser, and the private key never leaves it. On a phone it is kept in the platform’s secure storage (the Keychain on iOS, Keystore-backed storage on Android). In a browser it is kept in that browser’s local storage, which any script running on our page could read — a weaker home than a phone. Only the public key is uploaded, so others can encrypt messages to you.
Sending a message
- Your app fetches the recipient’s public key from our server.
- It generates a fresh random nonce for this one message.
- NaCl box encrypts the message with your private key + their public key + the nonce.
- The ciphertext and nonce are stored on the server; the plaintext never leaves your phone or browser.
- The recipient’s device reverses the process with their private key and your public key.
What our database stores per Secret chat message: sender id, conversation id, timestamp, a base64 ciphertext blob, and a nonce. What it can reveal about the content: nothing.
Honest limitations
E2E encryption protects message content, not metadata — we necessarily know who talked to whom and when, or the app couldn’t deliver messages.
Compare security codes. The first time, your app takes the other person’s public key from our server; it remembers that key on your device and, if it ever changes, warns you and holds your messages until you accept the new one (only while the app stays open: an unsent Secret message is never saved). Each Secret chat has a 60-digit security code, in 12 groups of 5: compare it with the other person in person or on a call. If it matches, the code on both devices was made from the same two keys. Until you compare, Secret chat relies on our server handing out the right key the first time, and accepting a new code without checking it hands what you send next to whoever set up that key. Messages sent before a change stay sealed to the old key.
Channels are not end-to-end encrypted. They are encrypted in transit and at rest, but your workspace admins can read them, and so could we if compelled by law. That is deliberate: a shared work channel is a business record, and companies need to be able to search, retain and hand over their own records. If a conversation should not be readable by your admin, use a Secret chat — that is exactly what the end-to-end encrypted half is for.
Secret chats have no recovery. Each person’s key lives only on their own device, so a new phone, a reinstall or another browser cannot open older secret messages — we cannot read them and we cannot restore them. (A recovery phrase existed for the old per-device scheme from September 2026; it was retired with that scheme in October.) There is no support backdoor.