docarray · jupyterjazz · May 15, 2023 · May 15, 2023 · May 15, 2023 · May 15, 2023
diff --git a/docarray/index/backends/in_memory.py b/docarray/index/backends/in_memory.py
@@ -39,15 +39,33 @@
 
 
 class InMemoryExactNNIndex(BaseDocIndex, Generic[TSchema]):
-    def __init__(self, docs: Optional[DocList] = None, **kwargs):
+    def __init__(
+        self,
+        docs: Optional[DocList] = None,
+        index_file_path: Optional[str] = None,
+        **kwargs,
+    ):
         """Initialize InMemoryExactNNIndex"""
         super().__init__(db_config=None, **kwargs)
         self._runtime_config = self.RuntimeConfig()
-        self._docs = (
-            docs
-            if docs is not None
-            else DocList.__class_getitem__(cast(Type[BaseDoc], self._schema))()
-        )
+
+        if docs and index_file_path:
+            raise ValueError(
+                'Initialize `InMemoryExactNNIndex` with either `docs` or '
+                '`index_file_path`, not both. Provide `docs` for a fresh index, or '
+                '`index_file_path` to use an existing file.'
+            )
+
+        if index_file_path:
+            self._docs = DocList.__class_getitem__(
+                cast(Type[BaseDoc], self._schema)
+            ).load_binary(file=index_file_path)
+        else:
+            self._docs = (
+                docs
+                if docs is not None
+                else DocList.__class_getitem__(cast(Type[BaseDoc], self._schema))()
+            )
 
     def python_type_to_db_type(self, python_type: Type) -> Any:
         """Map python type to database type.
@@ -293,3 +311,7 @@ def _text_search_batched(
         self, queries: Sequence[str], limit: int, search_field: str = ''
     ) -> _FindResultBatched:
         raise NotImplementedError(f'{type(self)} does not support text search.')
+
+    def persist(self, file: str = 'in_memory_index.bin') -> None:
+        """Persist InMemoryExactNNIndex into a binary file."""
+        self._docs.save_binary(file=file)
diff --git a/docs/user_guide/storing/index_in_memory.md b/docs/user_guide/storing/index_in_memory.md
@@ -33,6 +33,15 @@ doc_index.index(docs)
 doc_index = InMemoryExactNNIndex[MyDoc](docs)
 ```
 
+Additionally, you can preserve your index as a binary file and instantiate a new one using this file:
+```python
+# Save your existing index as a binary file
+doc_index.persist('docs.bin')
+
+# Initialize a new document index using the saved binary file
+new_doc_index = InMemoryExactNNIndex[MyDoc](index_file_path='docs.bin')
+```
+
 ## Configuration
 
 This section lays out the configurations and options that are specific to [InMemoryExactNNIndex][docarray.index.backends.in_memory.InMemoryExactNNIndex].

diff --git a/tests/index/in_memory/test_in_memory.py b/tests/index/in_memory/test_in_memory.py
@@ -120,3 +120,18 @@ def test_concatenated_queries(doc_index):
     docs, scores = doc_index.execute_query(q)
 
     assert len(docs) == 4
+
+
+def test_save_and_load(doc_index, tmpdir):
+    initial_num_docs = doc_index.num_docs()
+
+    binary_file = str(tmpdir / 'docs.bin')
+    doc_index.persist(binary_file)
+
+    new_doc_index = InMemoryExactNNIndex[SchemaDoc](index_file_path=binary_file)
+
+    docs, scores = new_doc_index.find(np.ones(10), search_field='tensor', limit=5)
+
+    assert len(docs) == 5
+    assert len(scores) == 5
+    assert new_doc_index.num_docs() == initial_num_docs