Network/hostname plan for exposing the proxy (LAN + external) #13
Notifications
Due Date
No due date set.
Blocks
#14 Author the docker-compose service for the chosen proxy
haylan/LLM-Server
Reference: haylan/LLM-Server#13
Reference in New Issue
Block a user
Part of #9
Question
Per #9's Notes, the proxy needs to be reachable LAN + external, mirroring Open WebUI's ai.home / ai.haylan.ch pattern via the existing NPM instance (see docs/network-access.md). What's the concrete plan: hostname(s), whether it shares Open WebUI's domain or gets its own subdomain, auth exposure concerns now that llama.cpp's raw (currently LAN-only, unauthenticated) API sits behind it, and any changes needed to llama.cpp's own network posture from map #1.
Resolved via grilling.
Hostnames: proxy.ai.home (LAN) / proxy.ai.haylan.ch (external) — new NPM Proxy Hosts, same pattern as Open WebUI's ai.home/ai.haylan.ch. LiteLLM's own virtual-key auth (per #12) is sufficient for external exposure — no extra NPM-level auth layer. LiteLLM's Admin UI (/ui) stays LAN-only, reachable only via proxy.ai.home — since it shares a port with the API by default, the external proxy.ai.haylan.ch NPM host needs an explicit rule denying the /ui path.
Documented in docs/network-access.md, committed on main.