fix: AmazonBedrockDocumentEmbedder to not modify Documents in place (#2174)

- Add dataclasses.replace import
- Update _embed_cohere to create new document instances
- Update _embed_titan to create new document instances
- Add immutability tests for both Cohere and Titan paths
- Update CHANGELOG.md with bug fix entry

Follows the pattern from FastEmbed, Optimum, and Nvidia integrations.

Author: GunaPalanivel

integrations/amazon_bedrock/CHANGELOG.md

@@ -1,12 +1,17 @@
 # Changelog
 
+## [unreleased]
+
+### 🐛 Bug Fixes
+
+- Fix `AmazonBedrockDocumentEmbedder` to not modify Documents in place (#2174)
+
 ## [integrations/amazon_bedrock-v6.0.0] - 2026-01-09
 
 ### 🧹 Chores
 
 - [**breaking**] Amazon_bedrock - drop Python 3.9 and use X|Y typing (#2685)
 
-
 ## [integrations/amazon_bedrock-v5.4.0] - 2026-01-08
 
 ### 🚀 Features
@@ -27,7 +32,6 @@
 
 - Relax model name validation for Bedrock Embedders (#2625)
 
-
 ## [integrations/amazon_bedrock-v5.3.0] - 2025-12-17
 
 ### 🚀 Features
@@ -57,7 +61,6 @@
 
 - S3Downloader - add `s3_key_generation_function` param to customize S3 key generation (#2343)
 
-
 ## [integrations/amazon_bedrock-v5.0.0] - 2025-09-22
 
 ### 🚀 Features
@@ -67,7 +70,6 @@
 
 ### 📚 Documentation
 
-
 ### 🧹 Chores
 
 - Bedrock - remove unused `stop_words` init parameter (#2275)
@@ -84,7 +86,6 @@
 - [**breaking**] Update AmazonBedrockChatGenerator to use the new fields in `StreamingChunk` (#2216)
 - [**breaking**] Use `ReasoningContent` to store reasoning content in `ChatMessage` instead of `ChatMessage.meta` (#2226)
 
-
 ### 🧹 Chores
 
 - Standardize readmes - part 2 (#2205)
@@ -100,7 +101,6 @@
 - Add framework name into UserAgent header for bedrock integration (#2168)
 - Standardize readmes - part 1 (#2202)
 
-
 ## [integrations/amazon_bedrock-v3.10.0] - 2025-08-06
 
 ### 🚀 Features
@@ -117,14 +117,12 @@
 
 - `AmazonBedrockChatGenerator` - fix bug with streaming + tool calls with no arguments (#2121)
 
-
 ## [integrations/amazon_bedrock-v3.9.0] - 2025-07-29
 
 ### 🚀 Features
 
 - Amazon Bedrock - multimodal support (#2114)
 
-
 ## [integrations/amazon_bedrock-v3.8.0] - 2025-07-04
 
 ### 🚀 Features
@@ -136,7 +134,6 @@
 - Remove black (#1985)
 - Improve typing for select_streaming_callback (#2008)
 
-
 ## [integrations/amazon_bedrock-v3.7.0] - 2025-06-11
 
 ### 🐛 Bug Fixes
@@ -154,21 +151,18 @@
 - Align core-integrations Hatch scripts (#1898)
 - Update md files for new hatch scripts (#1911)
 
-
 ## [integrations/amazon_bedrock-v3.6.2] - 2025-05-13
 
 ### 🧹 Chores
 
 - Extend error message for unknown model family in AmazonBedrockGenerator (#1733)
 
-
 ## [integrations/amazon_bedrock-v3.6.1] - 2025-05-13
 
 ### 🚜 Refactor
 
 - Add AmazonBedrockRanker and keep BedrockRanker as alias (#1732)
 
-
 ## [integrations/amazon_bedrock-v3.6.0] - 2025-05-09
 
 ### 🚀 Features
@@ -183,7 +177,6 @@
 
 - Refactor tests of AmazonBedrock Integration (#1671)
 
-
 ### 🧹 Chores
 
 - Update ChatGenerators with `deserialize_tools_or_toolset_inplace` (#1623)
@@ -205,7 +198,6 @@
 
 - AmazonBedrockGenerator - return response metadata (#1584)
 
-
 ### 🧪 Testing
 
 - Update tool serialization in tests to include `inputs_from_state` and `outputs_to_state` (#1581)
@@ -224,7 +216,6 @@
 
 - Review testing workflows (#1541)
 
-
 ## [integrations/amazon_bedrock-v3.2.1] - 2025-03-13
 
 ### 🐛 Bug Fixes
@@ -235,7 +226,6 @@
 
 - Update AWS Bedrock with improved docstrings and warning message (#1532)
 
-
 ### 🧹 Chores
 
 - Use Haystack logging across integrations (#1484)
@@ -246,14 +236,12 @@
 
 - Adding async to `AmazonChatGenerator` (#1445)
 
-
 ## [integrations/amazon_bedrock-v3.1.1] - 2025-02-26
 
 ### 🐛 Bug Fixes
 
 - Avoid thinking end tag on first content block (#1442)
 
-
 ## [integrations/amazon_bedrock-v3.1.0] - 2025-02-26
 
 ### 🚀 Features
@@ -275,26 +263,23 @@
 - Chore: Bedrock - manually fix changelog (#1319)
 - Fix error when empty document list (#1325)
 
-
 ## [integrations/amazon_bedrock-v3.0.0] - 2025-01-23
 
 ### 🚀 Features
 
-- *(AWS Bedrock)* Add Cohere Reranker (#1291)
+- _(AWS Bedrock)_ Add Cohere Reranker (#1291)
 - AmazonBedrockChatGenerator - add tools support (#1304)
 
 ### 🚜 Refactor
 
-- [**breaking**] AmazonBedrockGenerator - remove truncation  (#1314)
-
+- [**breaking**] AmazonBedrockGenerator - remove truncation (#1314)
 
 ## [integrations/amazon_bedrock-v2.1.3] - 2025-01-21
 
 ### 🧹 Chores
 
 - Bedrock - pin `transformers!=4.48.*` (#1306)
 
-
 ## [integrations/amazon_bedrock-v2.1.2] - 2025-01-20
 
 ### 🌀 Miscellaneous
@@ -307,27 +292,24 @@
 
 - Fixes to Bedrock Chat Generator for compatibility with the new ChatMessage (#1250)
 
-
 ## [integrations/amazon_bedrock-v2.1.0] - 2024-12-11
 
 ### 🚀 Features
 
 - Support model_arn in AmazonBedrockGenerator (#1244)
 
-
 ## [integrations/amazon_bedrock-v2.0.0] - 2024-12-10
 
 ### 🚀 Features
 
 - Update AmazonBedrockChatGenerator to use Converse API (BREAKING CHANGE) (#1219)
 
-
 ## [integrations/amazon_bedrock-v1.1.1] - 2024-12-03
 
 ### 🐛 Bug Fixes
 
 - AmazonBedrockChatGenerator with Claude raises moot warning for stream… (#1205)
-- Allow passing boto3 config to all  AWS Bedrock classes (#1166)
+- Allow passing boto3 config to all AWS Bedrock classes (#1166)
 
 ### 🧹 Chores
 
@@ -347,26 +329,23 @@
 
 - Adopt uv as installer (#1142)
 
-
 ## [integrations/amazon_bedrock-v1.0.5] - 2024-10-17
 
 ### 🚀 Features
 
 - Add prefixes to supported model patterns to allow cross region model ids (#1127)
 
-
 ## [integrations/amazon_bedrock-v1.0.4] - 2024-10-16
 
 ### 🐛 Bug Fixes
 
 - Avoid bedrock read timeout (add boto3_config param) (#1135)
 
-
 ## [integrations/amazon_bedrock-v1.0.3] - 2024-10-04
 
 ### 🐛 Bug Fixes
 
-- *(Bedrock)* Allow tools kwargs for AWS Bedrock Claude model (#976)
+- _(Bedrock)_ Allow tools kwargs for AWS Bedrock Claude model (#976)
 - Chat roles for model responses in chat generators (#1030)
 
 ### 🚜 Refactor
@@ -379,7 +358,7 @@
 
 ### 🌀 Miscellaneous
 
-- Modify regex to allow cross-region inference in bedrock  (#1120)
+- Modify regex to allow cross-region inference in bedrock (#1120)
 
 ## [integrations/amazon_bedrock-v1.0.1] - 2024-08-19
 
@@ -391,7 +370,6 @@
 
 - Normalising ChatGenerators output (#973)
 
-
 ## [integrations/amazon_bedrock-v1.0.0] - 2024-08-12
 
 ### 🚜 Refactor
@@ -402,7 +380,6 @@
 
 - Do not retry tests in `hatch run test` command (#954)
 
-
 ## [integrations/amazon_bedrock-v0.10.0] - 2024-08-12
 
 ### 🐛 Bug Fixes
@@ -466,7 +443,7 @@
 
 ### 🚀 Features
 
-- Add Mistral Amazon Bedrock support  (#632)
+- Add Mistral Amazon Bedrock support (#632)
 
 ### 📚 Documentation
 

integrations/amazon_bedrock/src/haystack_integrations/components/embedders/amazon_bedrock/document_embedder.py

@@ -1,4 +1,5 @@
 import json
+from dataclasses import replace
 from typing import Any
 
 from botocore.config import Config
@@ -186,10 +187,11 @@ def _embed_cohere(self, documents: list[Document]) -> list[Document]:
             )
             all_embeddings.extend(embeddings_list)
 
+        new_documents = []
         for doc, emb in zip(documents, all_embeddings, strict=True):
-            doc.embedding = emb
+            new_documents.append(replace(doc, embedding=emb))
 
-        return documents
+        return new_documents
 
     def _embed_titan(self, documents: list[Document]) -> list[Document]:
         """
@@ -214,10 +216,11 @@ def _embed_titan(self, documents: list[Document]) -> list[Document]:
             embedding = response_body["embedding"]
             all_embeddings.append(embedding)
 
+        new_documents = []
         for doc, emb in zip(documents, all_embeddings, strict=True):
-            doc.embedding = emb
+            new_documents.append(replace(doc, embedding=emb))
 
-        return documents
+        return new_documents
 
     @component.output_types(documents=list[Document])
     def run(self, documents: list[Document]) -> dict[str, list[Document]]:

integrations/amazon_bedrock/tests/test_document_embedder.py

@@ -257,6 +257,74 @@ def mock_invoke_model(*args, **kwargs):
             assert doc.content == docs[i].content
             assert doc.embedding == [0.1, 0.2, 0.3]
 
+    def test_run_cohere_does_not_modify_original_documents(self, mock_boto3_session):
+        embedder = AmazonBedrockDocumentEmbedder(model="cohere.embed-english-v3")
+
+        original_docs = [
+            Document(content="test 1", id="doc1"),
+            Document(content="test 2", id="doc2"),
+        ]
+
+        # Store original IDs to verify they're the same objects
+        original_doc_ids = [id(doc) for doc in original_docs]
+        original_embeddings = [doc.embedding for doc in original_docs]
+
+        with patch.object(embedder, "_client") as mock_client:
+            mock_client.invoke_model.return_value = {
+                "body": io.StringIO('{"embeddings": [[0.1, 0.2], [0.3, 0.4]]}'),
+            }
+
+            result = embedder.run(documents=original_docs)
+
+        # Verify originals are unchanged
+        assert all(doc.embedding is None for doc in original_docs)
+        assert original_embeddings == [None, None]
+
+        # Verify returned documents are NEW instances
+        returned_doc_ids = [id(doc) for doc in result["documents"]]
+        assert original_doc_ids != returned_doc_ids
+
+        # Verify returned documents have embeddings
+        assert result["documents"][0].embedding == [0.1, 0.2]
+        assert result["documents"][1].embedding == [0.3, 0.4]
+        assert result["documents"][0].content == "test 1"
+        assert result["documents"][1].content == "test 2"
+
+    def test_run_titan_does_not_modify_original_documents(self, mock_boto3_session):
+        embedder = AmazonBedrockDocumentEmbedder(model="amazon.titan-embed-text-v1")
+
+        original_docs = [
+            Document(content="test 1", id="doc1"),
+            Document(content="test 2", id="doc2"),
+        ]
+
+        # Store original IDs to verify they're the same objects
+        original_doc_ids = [id(doc) for doc in original_docs]
+        original_embeddings = [doc.embedding for doc in original_docs]
+
+        with patch.object(embedder, "_client") as mock_client:
+            # Titan returns one embedding at a time
+            mock_client.invoke_model.side_effect = [
+                {"body": io.StringIO('{"embedding": [0.1, 0.2]}')},
+                {"body": io.StringIO('{"embedding": [0.3, 0.4]}')},
+            ]
+
+            result = embedder.run(documents=original_docs)
+
+        # Verify originals are unchanged
+        assert all(doc.embedding is None for doc in original_docs)
+        assert original_embeddings == [None, None]
+
+        # Verify returned documents are NEW instances
+        returned_doc_ids = [id(doc) for doc in result["documents"]]
+        assert original_doc_ids != returned_doc_ids
+
+        # Verify returned documents have embeddings
+        assert result["documents"][0].embedding == [0.1, 0.2]
+        assert result["documents"][1].embedding == [0.3, 0.4]
+        assert result["documents"][0].content == "test 1"
+        assert result["documents"][1].content == "test 2"
+
     @pytest.mark.integration
     @pytest.mark.skipif(
         not os.getenv("AWS_ACCESS_KEY_ID")