Input one way, voice and lip-sync the other — each on its own lane.
Fast input and rich voice travel on independent lanes, each tuned to what it needs. Neither waits on the other.
Players can talk over a character and it reacts immediately, like a real conversation.
Voice arrives with matching mouth shapes, so characters look like they're actually speaking.
No. You keep Unreal / Unity / a custom server. We replace the realtime fabric layer — the thing that fans player input out to other players, broadcasts world ticks, and runs voice. Your game logic is untouched.
The relay is a transport, not a state authority. Your authoritative server still validates inputs and rejects them; we just deliver the bytes faster, with better tail latency and per-track auth. Anti-cheat systems integrate at the application layer the same way they do today.
Yes. QUIC datagrams (RFC 9221) are unreliable, unordered, single-packet — the equivalent of UDP-with-encryption. Use them for fast state. Use QUIC streams when you want reliable + in-order. Mix per-channel.
Photon and Mirror are application-layer netcode libraries — they assume UDP and roll their own reliability. GameLift is hosted UDP + matchmaking. We're the QUIC-shaped equivalent of all three transports underneath, with one wire, one auth model, and the same SDK your AI/voice/IoT/CRM stack uses. Less to operate.
On the public internet, single relay hop: caller → relay → callee adds ~10-25 ms over direct-RTT depending on geography. With multi-hop fanout (1 origin + 2 edge tiers), worst-case is one extra hop. Geographically-pinned relays keep p99 under 60 ms for most metros.
Telequick ships every modality on the same transport.