diff --git a/sidecar/app.py b/sidecar/app.py
index 19e15b3..132f9cd 100644
--- a/sidecar/app.py
+++ b/sidecar/app.py
@@ -19,7 +19,9 @@ from sidecar.manifest import load_manifest
 MANIFEST_PATH = os.getenv("MANIFEST_PATH", "/home/bigt/AI/llm/manifest.yaml")
 SIDECAR_PORT = int(os.getenv("SIDECAR_PORT", "8080"))
 LLAMA_SERVER_PORT = 8081
-LLAMA_STDERR_LOG = "/tmp/llama-server-stderr.log"
+LLAMA_STDERR_LOG = os.path.join(
+    os.path.dirname(MANIFEST_PATH), "llama-server-stderr.log"
+)
 
 # Global state
 _llama_server_process: Optional[asyncio.subprocess.Process] = None
@@ -128,6 +130,9 @@ async def _poll_llama_server_ready(max_retries: int = 60, interval: float = 0.5)
             pass
         await asyncio.sleep(interval)
 
+    # Flush and close the stderr handle so all data is on disk before we read
+    _close_stderr_log()
+
     # ── Dump stderr for diagnosis ──────────────────────────────────────
     print("llama-server did NOT become ready — dumping stderr:", flush=True)
     try: