diff --git a/build_store.py b/build_store.py
index fd2b477..551545c 100644
--- a/build_store.py
+++ b/build_store.py
@@ -125,8 +125,16 @@ def get_embed_tokenizer():
     global _EMBED_TOKENIZER
     if _EMBED_TOKENIZER is None:
         from transformers import AutoTokenizer
+        # model_max_length is raised because this tokenizer is used ONLY to
+        # count tokens for chunking -- the model never sees these sequences.
+        # Left at its default (512) it warns on every document longer than
+        # that ("Token indices sequence length is longer than..."), which is
+        # noise here, not a problem. Counts are unaffected either way.
         _EMBED_TOKENIZER = AutoTokenizer.from_pretrained(
-            EMBED_MODEL_NAME, cache_dir="./models", local_files_only=True
+            EMBED_MODEL_NAME,
+            cache_dir="./models",
+            local_files_only=True,
+            model_max_length=int(1e9),
         )
     return _EMBED_TOKENIZER
 
diff --git a/clippings_search/build_clippings.py b/clippings_search/build_clippings.py
index 90de713..1d1b369 100644
--- a/clippings_search/build_clippings.py
+++ b/clippings_search/build_clippings.py
@@ -73,8 +73,16 @@ def get_embed_tokenizer():
     global _EMBED_TOKENIZER
     if _EMBED_TOKENIZER is None:
         from transformers import AutoTokenizer
+        # model_max_length is raised because this tokenizer is used ONLY to
+        # count tokens for chunking -- the model never sees these sequences.
+        # Left at its default (512) it warns on every document longer than
+        # that ("Token indices sequence length is longer than..."), which is
+        # noise here, not a problem. Counts are unaffected either way.
         _EMBED_TOKENIZER = AutoTokenizer.from_pretrained(
-            EMBED_MODEL_NAME, cache_dir="./models", local_files_only=True
+            EMBED_MODEL_NAME,
+            cache_dir="./models",
+            local_files_only=True,
+            model_max_length=int(1e9),
         )
     return _EMBED_TOKENIZER