Thanks — I see you're hitting a recurring error:

Exception: worker connection closed unexpectedly

from the LiveKit agent worker. You also have a log line showing the worker registered successfully (wss://dpi-kccd98ec.livekit.cloud) right before the connection closed. I’ll help you diagnose likely causes and walk through targeted checks and mitigations. Quick summary of likely causes • Network instability causing the WebSocket/WebRTC connection to drop. • A firewall, VPN, or security proxy (for example, ZScaler or another DPI proxy) interfering with WebSocket, TURN, or encrypted media streams. • Intermediate proxies performing Deep Packet Inspection that break WebRTC/TURN traffic. Actionable checks and next steps 1. Run the Connection Test utility ◦ Use the LiveKit Connection Test to verify WebSocket, WebRTC, and TURN connectivity from the agent’s host. If tests fail for WebSocket or TURN, the problem is almost certainly network/proxy related. 1. Check for proxies/ DPI (ZScaler-like) ◦ If your network uses ZScaler or a DPI proxy, configure bypass rules for your LiveKit project domain and TURN servers. Example bypass entries: ▪︎ turn.livekit.cloud — detect_protocol(no) ▪︎ dpi-kccd98ec.livekit.cloud — detect_protocol(no) ◦ Bypassing protocol detection prevents the proxy from attempting to inspect or modify encrypted WebRTC/TURN traffic (which breaks connections). 1. Verify ports and protocols ◦ Ensure outbound ports and protocols required by WebSocket, WebRTC, and TURN are allowed from the agent host (including UDP for TURN if you use UDP). ◦ Temporarily disable VPNs or test from a network without corporate proxies to confirm if the proxy is the cause. 1. Inspect agent logs & retry behavior ◦ Confirm the agent’s retry/backoff is enabled. The agent log shows “failed to connect to livekit, retrying in 0s” which means it does attempt reconnects — keep an eye on repeated failures and timestamps to see if disconnections are transient or immediate after registration. 1. Reproduce from a different network/location ◦ Try running the worker from a different network (home network, cloud VM in a different region) to determine whether the issue is local-network specific. What I can help with next • If you can share the outcome of the Connection Test or the specific network environment (behind ZScaler or other proxy?), I’ll suggest the exact bypass rules and further steps. • If you prefer, paste a sanitized portion of more detailed agent logs (timestamps around registration and the failure) and I’ll help pinpoint whether the disconnect is immediately post-registration or after some activity. If you don’t use a proxy or the Connection Test passes, tell me the test results and I’ll guide you through additional logs and configuration checks. Sources: Configure ZScaler proxy for LiveKit connections | LiveKit | Diagnosing Connection Errors with the Connection Test Utility | LiveKit | server-sdk-go | livekit/livekit_composite | DeepWiki