EVIDENCE BEFORE PROMISES

What we tested.
What still needs a call.

A public beta should be clear about its limits. This is an engineering verification summary, not an independent security audit or uptime guarantee.

Last reviewed: .

Browser workflows checked.

Chromium browser checks covered encrypted contact backup export and restore, clearing local app data, loading the cached Phone Book offline and temporary contacts that disappear after reload. Emulated layouts were checked at phone and tablet widths. Emulation does not prove physical iPhone or Android compatibility.

Private signaling and worker updates.

34 automated application tests passed in the recorded verification run. A production-origin test kept a signaling connection open during a worker deployment, joined another peer to the same room afterwards, checked authenticated relay-credential routing and exercised rollback. This verifies the signaling path, not uninterrupted audio on every network.

Existing rooms stay with their original worker while new rooms use the new version. Gateway restarts, host failures and incompatible protocol changes remain separate availability risks.

TURN is configured; physical audio testing remains.

The public deployment reported a configured relay, and the server successfully obtained temporary Cloudflare TURN credentials. That verifies configuration and credential generation. It does not by itself verify media transport on Apple, Pixel, restrictive Wi-Fi or a particular mobile operator.

Before accepting a paid pilot, test two-way audio on the customer's actual devices and separate networks, permission recovery, a longer call, reconnection and the chosen weak-network profile.

What the weak-network test means.

An encrypted Tiny-message browser test passed with application-message pacing at 6 kbps, added send delay and a deliberately lost first receipt. No duplicate message or microphone capture was observed in that test. This was a controlled application-level test, not a real 2G field trial or a measurement of total traffic.

The lowest voice setting requests a 6 kbps codec target; headers, encryption and relay transport add overhead. Read the weak-connection guide →

Security and privacy boundaries.

No independent penetration test or external security certification is claimed. Application code does not maintain a durable contact book, recording store or call-history database on the server. Infrastructure providers can still process metadata, and another participant can record a conversation.

We do not promise untraceability, guaranteed voice anonymity or forensic erasure. Read the privacy explanation →

Help verify your setup.

Share the browser and operating-system versions, network type, steps and the error shown. Never include private invitations, passwords, TURN tokens, recordings or contact backups in a report.

Email a compatibility report ↗

For a deployment on your own domain, start with a scoped pilot.