bryan/nixos
0
0
Fork 0
mirror of https://github.com/itme-brain/nixos.git synced 2026-05-08 14:50:12 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

seperate routing for llama & stack

Browse source
This commit is contained in:
Bryan Ramos 2026-04-14 00:41:39 -04:00
parent 3feb5ddc6b
commit 46adf8e9f0
1 changed files with 5 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

9
system/machines/server/modules/nginx/default.nix
View file

@ -140,15 +140,16 @@ in
useACMEHost = domain; useACMEHost = domain;
forceSSL = true; forceSSL = true;
# Web UI — llama.cpp chat interface (browser) # Web UI + llama.cpp API (browser, /v1/* calls from the UI)
# Auth handled by llama.cpp itself (--api-key flag) # Auth handled by llama.cpp itself (--api-key flag)
locations."/" = { locations."/" = {
proxyPass = "http://192.168.0.23:8000"; proxyPass = "http://192.168.0.23:8000";
proxyWebsockets = true; proxyWebsockets = true;
}; };
# API — Llama Stack (opencode, programmatic clients) # Llama Stack API (opencode, programmatic clients)
locations."/v1/" = { # Clients use baseURL: https://ai.ramos.codes/stack/v1
locations."/stack/v1/" = {
proxyPass = "http://192.168.0.23:8321/v1/"; proxyPass = "http://192.168.0.23:8321/v1/";
proxyWebsockets = true; proxyWebsockets = true;
extraConfig = apiKeyAuth + '' extraConfig = apiKeyAuth + ''
@ -158,7 +159,7 @@ in
}; };
# Llama Stack beta API # Llama Stack beta API
locations."/v1beta/" = { locations."/stack/v1beta/" = {
proxyPass = "http://192.168.0.23:8321/v1beta/"; proxyPass = "http://192.168.0.23:8321/v1beta/";
proxyWebsockets = true; proxyWebsockets = true;
extraConfig = apiKeyAuth + '' extraConfig = apiKeyAuth + ''