Run prod agent on gpt-5.4 via OpenAI fallback

Production only has OPENAI_API_KEY configured, so the blueprint-synced
LLM_MODEL=claude-fable-5 made every model call fail with 'Could not
resolve authentication method'. Point LLM_MODEL at gpt-5.4 (the
pre-migration prod model) to route through ChatOpenAI instead.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: jerelvelarde

render.yaml

@@ -23,8 +23,10 @@ services:
         sync: false
       - key: LANGSMITH_API_KEY
         sync: false
+      # gpt-* routes to OpenAI (prod only has OPENAI_API_KEY); set back to
+      # claude-fable-5 once ANTHROPIC_API_KEY is configured on the service
       - key: LLM_MODEL
-        value: claude-fable-5
+        value: gpt-5.4-2026-03-05
     buildFilter:
       paths:
         - apps/agent/**