Summary

I'm trying to secure Clawdbot with Pomerium (an Identity Aware Proxy). Full disclosure, I work at Pomerium. The proxy handles authentication, so Clawdbot doesn't need its own auth.

But when I bind to LAN, Clawdbot forces token authentication. This breaks WebSocket connections because browsers can't pass the token in WebSocket message payloads. The web UI loads but shows "disconnected (1008): unauthorized" for all real-time features.

Proposed solution

Allow gateway.auth: "off" when gateway.bind: "lan" for reverse proxy scenarios.

Could add an explicit opt-in flag like:

{
  gateway: {
    bind: "lan",
    auth: {
      mode: "off",
      allowUnauthenticatedProxy: true
    }
  }
}

This would only work when users explicitly opt in and understand they need a reverse proxy handling auth.

Alternatives considered

1. iptables forwarding to keep loopback binding - ran into complex networking issues with Kubernetes service routing
2. Passing token via Pomerium headers - doesn't work because the token is required in WebSocket message data, not HTTP headers
3. Just accepting the errors - web UI is non-functional without WebSockets
4. Using Docker instead of Kubernetes - works fine with simple Docker setups (host networking or host.docker.internal), but not suitable for Kubernetes deployments where service routing is more complex.

Additional context

My setup: Kubernetes with Pomerium securing services in my cluster. Clawdbot runs on the host, not exposed to the internet. Only Pomerium can reach it.

This is a similar pattern to how people run apps behind Caddy with authentication or nginx with OAuth plugins.