fix(infra): remove buffer/compress from llm-api for real SSE streaming

- llm-api uses llm-chain instead of common-chain - llm-chain excludes buffer & compress middlewares - Prevents Traefik from buffering streaming SSE responses - Each token is now flushed immediately
2026-07-26 19:14:52 +07:00
parent 43f0df4493
commit be7348a27d
2 changed files with 9 additions and 1 deletions
@@ -61,6 +61,14 @@ http:
            - "^/wp-login\\.php"
            - "^/config\\.php"

+    # ── LLM Stream Chain (no buffer/compress — SSE needs real-time) ──
+    llm-chain:
+      chain:
+        middlewares:
+          - secure-headers
+          - retry
+          - rate-limit
+
    # ── Common Chain ──
    common-chain:
      chain: