diff --git a/aucourt_ingest/main.py b/aucourt_ingest/main.py
index 0accdf6..15e3ce9 100644
--- a/aucourt_ingest/main.py
+++ b/aucourt_ingest/main.py
@@ -319,14 +319,8 @@ def cmd_serve(args):
             database=config.storage.neo4j_database,
         )
 
-    # VectorIndex requires ChromaDB directory — skip for memory backend in tests
-    vector_index = None
-    if backend != "memory":
-        from aucourt_ingest.storage.vector_index import VectorIndex
-        vector_index = VectorIndex(str(config.storage.chromadb_dir))
-    else:
-        from aucourt_ingest.storage.vector_index import VectorIndex
-        vector_index = VectorIndex(str(config.storage.chromadb_dir))
+    from aucourt_ingest.storage.vector_index import VectorIndex
+    vector_index = VectorIndex(str(config.storage.chromadb_dir))
 
     from aucourt_ingest.api.app import create_app
     app = create_app(graph_db, vector_index, max_tokens)
@@ -336,10 +330,7 @@ def cmd_serve(args):
     uvicorn.run(app, host=host, port=port)
 
 
-async def async_main():
-    parser = build_parser()
-    args = parser.parse_args()
-
+async def async_main(args):
     if not args.mode:
         parser.print_help()
         return
@@ -354,13 +345,18 @@ async def async_main():
         await cmd_audit(args)
     elif args.mode == "process":
         await cmd_process(args)
-    elif args.mode == "serve":
-        cmd_serve(args)
 
 
 def main():
     logging.basicConfig(level=logging.INFO, format="%(asctime)s %(name)s %(levelname)s %(message)s")
-    asyncio.run(async_main())
+    parser = build_parser()
+    args = parser.parse_args()
+
+    if args.mode == "serve":
+        cmd_serve(args)
+        return
+
+    asyncio.run(async_main(args))
 
 
 if __name__ == "__main__":