diff --git a/src/raglite/_insert.py b/src/raglite/_insert.py
index bd63c29e..91a4c4fd 100644
--- a/src/raglite/_insert.py
+++ b/src/raglite/_insert.py
@@ -1,5 +1,6 @@
 """Index documents."""
 
+import os
 from concurrent.futures import ThreadPoolExecutor, as_completed
 from contextlib import nullcontext
 from functools import partial
@@ -89,7 +90,11 @@ def _create_chunk_records(
     return document, chunk_records, chunk_embedding_records_list
 
 
-def insert_documents(  # noqa: C901
+# Cap the number of worker threads to avoid excessive resource usage.
+MAX_DEFAULT_WORKERS = 4  # Prevents oversubscription and high memory usage.
+
+
+def insert_documents(  # noqa: C901, PLR0912
     documents: list[Document],
     *,
     max_workers: int | None = None,
@@ -130,6 +135,11 @@ def insert_documents(  # noqa: C901
         documents = [doc for doc in documents if doc.id not in existing_doc_ids]
         if not documents:
             return
+
+    # Heuristic based on cpu count, amount of documents, and a cap of 4.
+    if max_workers is None:
+        max_workers = min(os.cpu_count() or 1, len(documents), MAX_DEFAULT_WORKERS)
+
     # For DuckDB databases, acquire a lock on the database.
     if engine.dialect.name == "duckdb":
         db_url = make_url(config.db_url) if isinstance(config.db_url, str) else config.db_url