fix bedrock with file attachments (#45)

Author: cosminacho

packages/uipath_langchain_client/CHANGELOG.md

@@ -2,6 +2,11 @@
 
 All notable changes to `uipath_langchain_client` will be documented in this file.
 
+## [1.5.4] - 2026-03-19
+
+### Fix
+- Fix bedrock clients with file attachments
+
 ## [1.5.3] - 2026-03-18
 
 ### Fix

packages/uipath_langchain_client/src/uipath_langchain_client/__version__.py

@@ -1,3 +1,3 @@
 __title__ = "UiPath LangChain Client"
 __description__ = "A Python client for interacting with UiPath's LLM services via LangChain."
-__version__ = "1.5.3"
+__version__ = "1.5.4"

packages/uipath_langchain_client/src/uipath_langchain_client/clients/bedrock/chat_models.py

@@ -15,9 +15,32 @@
 try:
     from anthropic import AnthropicBedrock, AsyncAnthropicBedrock
     from langchain_aws.chat_models import ChatBedrock, ChatBedrockConverse
+    from langchain_aws.chat_models import bedrock as _bedrock_module
     from langchain_aws.chat_models.anthropic import ChatAnthropicBedrock
 
     from uipath_langchain_client.clients.bedrock.utils import WrappedBotoClient
+
+    _original_format_data_content_block = _bedrock_module._format_data_content_block
+
+    def _patched_format_data_content_block(block: dict) -> dict:
+        """Extended version that also handles file/document blocks for Anthropic API."""
+        if block["type"] == "file":
+            if "base64" not in block and block.get("source_type") != "base64":
+                raise ValueError("File data only supported through in-line base64 format.")
+            if "mime_type" not in block:
+                raise ValueError("mime_type key is required for base64 data.")
+            return {
+                "type": "document",
+                "source": {
+                    "type": "base64",
+                    "media_type": block["mime_type"],
+                    "data": block.get("base64") or block.get("data", ""),
+                },
+            }
+        return _original_format_data_content_block(block)
+
+    _bedrock_module._format_data_content_block = _patched_format_data_content_block
+
 except ImportError as e:
     raise ImportError(
         "The 'aws' extra is required to use UiPathBedrockChatModel and UiPathBedrockChatModelConverse. "

packages/uipath_langchain_client/src/uipath_langchain_client/clients/bedrock/utils.py

@@ -29,6 +29,22 @@ def __init__(self, region_name: str = "PLACEHOLDER"):
         self.events = _MockEventHooks()
 
 
+def _serialize_bytes(obj: Any) -> Any:
+    """Recursively encode bytes values to base64 strings for JSON serialization.
+
+    This mimics boto3's serializer which re-encodes bytes to base64 before
+    sending as JSON. Needed because LangChain's ChatBedrockConverse decodes
+    base64 content (images, PDFs) into raw bytes objects.
+    """
+    if isinstance(obj, bytes):
+        return base64.b64encode(obj).decode("utf-8")
+    if isinstance(obj, dict):
+        return {k: _serialize_bytes(v) for k, v in obj.items()}
+    if isinstance(obj, list):
+        return [_serialize_bytes(item) for item in obj]
+    return obj
+
+
 class WrappedBotoClient:
     def __init__(self, httpx_client: Client | None = None, region_name: str = "PLACEHOLDER"):
         self.httpx_client = httpx_client
@@ -37,7 +53,7 @@ def __init__(self, httpx_client: Client | None = None, region_name: str = "PLACE
     def _stream_generator(self, request_body: dict[str, Any]) -> Iterator[dict[str, Any]]:
         if self.httpx_client is None:
             raise ValueError("httpx_client is not set")
-        with self.httpx_client.stream("POST", "/", json=request_body) as response:
+        with self.httpx_client.stream("POST", "/", json=_serialize_bytes(request_body)) as response:
             buffer = EventStreamBuffer()
             for chunk in response.iter_bytes():
                 buffer.add_data(chunk)
@@ -69,11 +85,13 @@ def converse(
             raise ValueError("httpx_client is not set")
         return self.httpx_client.post(
             "/",
-            json={
-                "messages": messages,
-                "system": system,
-                **params,
-            },
+            json=_serialize_bytes(
+                {
+                    "messages": messages,
+                    "system": system,
+                    **params,
+                }
+            ),
         ).json()
 
     def converse_stream(