How Tabella's Encryption Actually Works
Last updated: July 2026
We didn't invent the cryptography that protects your messages, and we're not going to pretend we did. This page explains exactly what Tabella uses, where it came from, who has checked it, and what we genuinely don't know.
Why this page exists
Most messaging apps ask you to trust a marketing claim: "military-grade encryption," "unbreakable," "the most private app." We think that's the wrong way to earn trust. We'd rather tell you the specific, checkable facts and let you draw your own conclusion — including the parts that aren't flattering.
What Tabella actually uses
Tabella is built on the Matrix protocol, an open communications standard. The cryptography underneath it — the part that actually encrypts your messages — is called Olm and Megolm, implemented in a Rust library called vodozemac. Tabella's apps run this directly: when the app starts, it explicitly turns on this cryptography engine before anything else happens. We did not write our own encryption. We picked a well-established one and built on top of it, which is the responsible way to do this — cryptography is one of the few areas of software where "we rolled our own" is a warning sign, not a selling point.
Is this the same thing Signal uses?
Related, but not identical — and we want to be precise about the difference rather than let the similarity imply more than it should.
Both are built on the same core idea, called the Double Ratchet: a way for two devices to keep generating new encryption keys for every message, so that even if one key were ever compromised, past and future messages stay safe. Matrix's Olm protocol was explicitly designed by following Signal's approach closely. But they are separate implementations, maintained by separate teams, and there is no single study that tests them head-to-head and declares one better. Each has been evaluated on its own:
- The Signal Protocol has a published academic security proof from cryptography researchers at Oxford, Queensland University of Technology, and McMaster University (2016), which found it cryptographically sound.
- vodozemac has an independent security audit, described below.
The two implementations have a few known, documented differences. One example: Matrix's Olm protocol uses a shorter authentication tag (a piece of data that detects tampering) than Signal's implementation does — a real, acknowledged design trade-off from Matrix's early history, not something we're glossing over.
The independent audit
In 2022, Matrix.org commissioned Least Authority, a security firm specializing in audits of privacy-critical software, to audit vodozemac. The audit found 10 issues worth addressing. 8 were fixed during the audit itself. 2 were judged lower-priority and left on an ongoing maintenance list. The full report is public — we link it below rather than just asserting it exists.
Security research on vodozemac hasn't stopped since 2022, and that's a good thing, not a red flag. Independent researchers periodically publish new findings, and the Matrix security team publishes public, technical responses to each one — including one as recently as February 2026, which the team addressed by publicly explaining why the specific concern raised wasn't practically exploitable under Matrix's design, while still agreeing to add an extra defensive check in a future release. An actively-scrutinized cryptography library with a team that responds in public is exactly what you want. One that goes quiet after its first audit is the one to worry about.
What we don't claim
- We don't claim Tabella's encryption has been tested "better" than any specific competitor's. Different libraries, different audits, not a horse race.
- We don't claim to have commissioned the vodozemac audit ourselves — Matrix.org did, because vodozemac is their project, not ours. We use it; we didn't pay for or run that audit.
- We don't claim every patch and improvement since 2022 has been independently re-audited from scratch. Software evolves; audits are a snapshot, not a permanent certificate.
- We don't claim every byte on our servers is unreadable to us — only that your normal, person-to-person messages, calls, and files are. Two opt-in features are not end-to-end encrypted by their nature, and we say so plainly: programmatic sends from a bot or integration you connect (they hand our server a message to relay), and hosted AI transcription for organizations that explicitly sign a business agreement and turn it on. If you never enable those, your server-side data is ciphertext we cannot read — full stop.
Our approach, in one sentence
Use cryptography that has been genuinely, publicly scrutinized; say exactly what that means and doesn't mean; and treat "we're not lying to you" as a stronger promise than any adjective we could put in front of the word "encryption."
Sources
- Matrix.org — Independent public audit of vodozemac (May 2022)
- Least Authority — vodozemac Final Audit Report (PDF, 30 March 2022)
- Matrix.org — Analysis of reported issues in vodozemac (February 2026)
- Cohn-Gordon et al. — A Formal Security Analysis of the Signal Messaging Protocol (2016)
- vodozemac source code (matrix-org/vodozemac on GitHub)