<50ms
Sub-50ms latency
Region-routed SFU architecture with median RTT under 50ms across all shards.
Technology
Under the hood
The technical details behind Lag's voice infrastructure. Region-routed architecture, redundant audio encoding, and performance metrics from real deployments.
Performance snapshot
<50ms
Median RTT
Region-routed baseline
1.3s
P95 Join Time
Room connect path
99.2%
Packet Recovery
Redundant audio encoding
Opus 48kHz
Audio Codec
Low-latency standard
3,000+
Room Capacity
Concurrent listeners per node
3 layers
Adaptive Quality
Auto-adjusts to connection
99.2%
99.2% packet recovery
Redundant audio encoding recovers dropped packets before the codec notices.
<1%
<1% CPU overhead
Lean runtime designed to stay out of the way. No Electron bloat.
Full stack
Self-host ready
Full Terraform IaC, Docker images, and deployment docs. Own your voice infra.
Infrastructure
Designed to be deployed, not just demoed.
We cut social complexity so the engineering stays focused on what matters: low-latency comms, stability under load, and clean deployments.
Explore self-hosting Self-host
Open deployment
Terraform configs, Docker images, and rollout docs included. Deploy on your metal.
Headless
CLI-first workflow
Voice chat from your terminal. Scriptable, pipeable, no GUI required.
Private
Zero telemetry
No tracking pixels, no analytics beacons, no data leaves your deployment.
3 layers
Adaptive streaming
3-layer simulcast with auto-adjustment. Opus 48kHz codec standard.
Core principles
Every decision optimizes for the call.
Region-routed voice shards, redundant audio encoding, adaptive simulcast, and full self-host support from day one. No feature exists unless it makes voice better.
Read our commitmentDesign principles
Latency first
Prioritize low delay and predictable room joins during active sessions.
No bloat
Focus engineering on comms quality and reliability, not social noise.
Security posture
Design decisions favor access control and lower data exposure.