LingXizhi's AI capabilities are provided by a local Rust kernel. The kernel listens on 127.0.0.1:3721 by default — bound to the local loopback address only, never exposed externally; if the port is occupied it auto-increments (up to 10 attempts) and writes the result back to configuration. At startup the kernel prints the actual port and the access token, and the IDE calls the following HTTP endpoints with that token.
Login-state-related cloud APIs are proxied uniformly by the kernel: /local/auth/* (login, register, send verification code, reset password), /local/plans, /local/llm-models and /local/llm-vendors are public endpoints; /local/subscription, /local/usage, /local/keys, /local/user-llm-config, /local/rush/* and others are protected by the License gate.
!
Local calls only: the kernel listens on 127.0.0.1 only — do not expose it to the LAN via firewall rules or port forwarding, and do not kill the kernel process manually; AI chat, completion and quota queries would all break. Restart the IDE to recover.
✦
First troubleshooting step: when AI features stop responding, run curl http://127.0.0.1:3721/health to check whether the kernel is ready; connection failures are usually fixed by restarting the IDE. See the FAQ for more issues.