LiveKit

Recently shipped Agents v1.7.0 with PII redaction and expressive-mode voice, plus Egress v1.14.0's rewritten sync engine and v1.14.1 resilience fixes.

Agents v1.7.0 added PII redaction for Agent Observability. Detected entities are semantically redacted from chat history and audio recordings, and sensitive data is filtered from logs and traces while retaining diagnostic context. Sensitive trace attributes and log fields were renamed — third-party observability queries on those names need updating.

Expressive mode lets voice agents speak with natural prosody and emotion. Speech delivery follows emotion tags generated from conversation context, enabled in AgentSession.

Egress v1.14.0 rewrote A/V synchronization and added the v2 API. Per-track drift probes and audio tempo control replace the old sync engine, a two-tiered pacer smooths output, and sustained high CPU now drains egress instead of hard-killing it. v1.14.1 fixed an ImageSink deadlock that wedged the pipeline on image upload failure, added upload queue headroom for frame bursts, and stopped HLS audio drops.

Rust SDKs fixed a security bug in NativeVideoSource keepalive frames. webrtc::I420Buffer::Create left pixel planes uninitialized, so pre-capture keepalive frames could leak recycled heap contents to subscribers as first keyframes — now they emit black frames. NVIDIA NVENC AV1 encoding is available when the GPU supports it.

Server v1.13.6 added WARP support and new limits. Experimental WARP (SPED + SNAP) is negotiated with the SFU, plus per-participant TURN allocation quotas, pending-track limits, and API request body size caps. H.264 baseline was removed from default enabled codecs. CLI v2.18.3 added --export for simulated runs and richer deploy output.