fix: chats keep failing on Anthropic and Bedrock after a tool call is saved incorrectly (#31431)

Sometimes a reply where the model used tools gets saved with a tool result that no call in that reply asked for, or with a tool call that never got its result. The chat history was then sent to the provider unchanged, Anthropic and Bedrock rejected it, and every following message in that chat failed until the user deleted the broken reply. Now each tool call is only kept together with its own result from the same reply, and the unmatched calls and results are left out of what gets sent to the model. The chat itself is not changed, and correctly saved chats are sent exactly as before.

Fixes #28937

backend/open_webui/utils/middleware.py

@@ -115,6 +115,7 @@ from open_webui.utils.misc import (
     get_last_user_message_item,
     get_message_list,
     get_output_text,
+    get_paired_tool_call_ids,
     get_response_error_detail,
     get_reasoning_details,
     get_system_message,
@@ -2306,25 +2307,12 @@ def process_messages_with_output(
 
 
 def sanitize_tool_pairs(messages: list[dict]) -> list[dict]:
-    tool_result_ids = {
-        message.get('tool_call_id')
-        for message in messages
-        if message.get('role') == 'tool' and message.get('tool_call_id')
-    }
-
-    tool_call_ids = {
-        tool_call.get('id')
-        for message in messages
-        for tool_call in (message.get('tool_calls') or [])
-        if message.get('role') == 'assistant' and tool_call.get('id')
-    }
+    paired_ids_by_message = get_paired_tool_call_ids(messages)
 
     sanitized = []
-    for message in messages:
+    for message, paired_ids in zip(messages, paired_ids_by_message):
         if message.get('role') == 'assistant' and message.get('tool_calls'):
-            kept = [
-                tool_call for tool_call in message.get('tool_calls') or [] if tool_call.get('id') in tool_result_ids
-            ]
+            kept = [tool_call for tool_call in message.get('tool_calls') or [] if tool_call.get('id') in paired_ids]
             if kept:
                 sanitized.append({**message, 'tool_calls': kept})
             else:
@@ -2333,7 +2321,7 @@ def sanitize_tool_pairs(messages: list[dict]) -> list[dict]:
                 clean.pop('reasoning_items', None)
                 if clean.get('content'):
                     sanitized.append(clean)
-        elif message.get('role') != 'tool' or message.get('tool_call_id') in tool_call_ids:
+        elif message.get('role') != 'tool' or message.get('tool_call_id') in paired_ids:
             sanitized.append(message)
 
     return sanitized

backend/open_webui/utils/misc.py

@@ -261,6 +261,26 @@ def get_output_text(output: list | None) -> str:
     return '\n'.join(texts)
 
 
+def get_paired_tool_call_ids(messages: list[dict]) -> list[set[str]]:
+    """Tool call ids answered within each message's assistant-plus-tool-results block."""
+    paired_ids_by_message = [set() for _ in messages]
+    for index, message in enumerate(messages):
+        if message.get('role') != 'assistant' or not message.get('tool_calls'):
+            continue
+
+        block_end = index + 1
+        while block_end < len(messages) and messages[block_end].get('role') == 'tool':
+            block_end += 1
+
+        requested_ids = {tool_call.get('id') for tool_call in message['tool_calls'] if tool_call.get('id')}
+        completed_ids = {tool_message.get('tool_call_id') for tool_message in messages[index + 1 : block_end]}
+        paired_ids = requested_ids & completed_ids
+        for block_index in range(index, block_end):
+            paired_ids_by_message[block_index] = paired_ids
+
+    return paired_ids_by_message
+
+
 def reconcile_tool_pairs(messages: list[dict]) -> list[dict]:
     """Drop unpaired tool_use / tool_result from a reconstructed conversation.
 
@@ -271,22 +291,14 @@ def reconcile_tool_pairs(messages: list[dict]) -> list[dict]:
 
     Well-formed output is unaffected: every id pairs, so nothing is stripped.
     """
-    completed_tool_call_ids = {
-        message['tool_call_id'] for message in messages if message.get('role') == 'tool' and message.get('tool_call_id')
-    }
-    requested_tool_call_ids = {
-        tool_call['id']
-        for message in messages
-        for tool_call in message.get('tool_calls') or ()
-        if message.get('role') == 'assistant' and tool_call.get('id')
-    }
+    paired_ids_by_message = get_paired_tool_call_ids(messages)
 
     reconciled_messages = []
-    for message in messages:
+    for message, paired_ids in zip(messages, paired_ids_by_message):
         role = message.get('role')
 
-        # Orphan tool result — no assistant ever claimed this call_id.
-        if role == 'tool' and message.get('tool_call_id') not in requested_tool_call_ids:
+        # Orphan tool result: not claimed by the assistant heading its tool block.
+        if role == 'tool' and message.get('tool_call_id') not in paired_ids:
             continue
 
         # Non-assistant or no tool_calls — pass through unchanged.
@@ -294,10 +306,8 @@ def reconcile_tool_pairs(messages: list[dict]) -> list[dict]:
             reconciled_messages.append(message)
             continue
 
-        # Keep only tool_calls whose id received a tool-role response.
-        valid_tool_calls = [
-            tool_call for tool_call in message['tool_calls'] if tool_call.get('id') in completed_tool_call_ids
-        ]
+        # Keep only tool_calls whose id received an adjacent tool-role response.
+        valid_tool_calls = [tool_call for tool_call in message['tool_calls'] if tool_call.get('id') in paired_ids]
 
         if valid_tool_calls:
             reconciled_messages.append({**message, 'tool_calls': valid_tool_calls})