feat(ai): Add original input length meta attribute

sentry_sdk/ai/utils.py

@@ -14,6 +14,7 @@
 
 import sentry_sdk
 from sentry_sdk.utils import logger
+from sentry_sdk.consts import SPANDATA
 
 MAX_GEN_AI_MESSAGE_BYTES = 20_000  # 20KB
 # Maximum characters when only a single message is left after bytes truncation
@@ -698,6 +699,8 @@ def truncate_and_annotate_messages(
     if len(messages) > 1:
         scope._gen_ai_original_message_count[span.span_id] = len(messages)
 
+    span.set_data(SPANDATA.META_GEN_AI_ORIGINAL_INPUT_MESSAGES_LENGTH, len(messages))
+
     return [truncated_message]
 
 

sentry_sdk/consts.py

@@ -869,6 +869,14 @@ class SPANDATA:
     Example: "a1b2c3d4e5f6"
     """
 
+    META_GEN_AI_ORIGINAL_INPUT_MESSAGES_LENGTH = (
+        "sentry.sdk_meta.gen_ai.input.messages.original_length"
+    )
+    """
+    The original number of input non-system instruction messages, before SDK trimming.
+    Example: 4
+    """
+
 
 class SPANSTATUS:
     """

tests/integrations/anthropic/test_anthropic.py

@@ -1010,6 +1010,7 @@ def test_anthropic_message_truncation(sentry_init, capture_events):
     assert len(parsed_messages) == 1
     assert "small message 5" in str(parsed_messages[0])
 
+    assert chat_span["data"][SPANDATA.META_GEN_AI_ORIGINAL_INPUT_MESSAGES_LENGTH] == 5
     assert tx["_meta"]["spans"]["0"]["data"]["gen_ai.request.messages"][""]["len"] == 5
 
 
@@ -1061,6 +1062,7 @@ async def test_anthropic_message_truncation_async(sentry_init, capture_events):
     assert len(parsed_messages) == 1
     assert "small message 5" in str(parsed_messages[0])
 
+    assert chat_span["data"][SPANDATA.META_GEN_AI_ORIGINAL_INPUT_MESSAGES_LENGTH] == 5
     assert tx["_meta"]["spans"]["0"]["data"]["gen_ai.request.messages"][""]["len"] == 5
 
 

tests/integrations/google_genai/test_google_genai.py

@@ -990,6 +990,7 @@ def test_google_genai_message_truncation(
     assert parsed_messages[0]["role"] == "user"
     assert small_content in parsed_messages[0]["content"]
 
+    assert invoke_span["data"][SPANDATA.META_GEN_AI_ORIGINAL_INPUT_MESSAGES_LENGTH] == 2
     assert (
         event["_meta"]["spans"]["0"]["data"]["gen_ai.request.messages"][""]["len"] == 2
     )

tests/integrations/langchain/test_langchain.py

@@ -1070,6 +1070,8 @@ def test_langchain_message_truncation(sentry_init, capture_events):
     assert isinstance(parsed_messages, list)
     assert len(parsed_messages) == 1
     assert "small message 5" in str(parsed_messages[0])
+
+    assert llm_span["data"][SPANDATA.META_GEN_AI_ORIGINAL_INPUT_MESSAGES_LENGTH] == 5
     assert tx["_meta"]["spans"]["0"]["data"]["gen_ai.request.messages"][""]["len"] == 5
 
 

tests/integrations/langgraph/test_langgraph.py

@@ -1384,4 +1384,6 @@ def original_invoke(self, *args, **kwargs):
     assert isinstance(parsed_messages, list)
     assert len(parsed_messages) == 1
     assert "small message 5" in str(parsed_messages[0])
+
+    assert invoke_span["data"][SPANDATA.META_GEN_AI_ORIGINAL_INPUT_MESSAGES_LENGTH] == 5
     assert tx["_meta"]["spans"]["0"]["data"]["gen_ai.request.messages"][""]["len"] == 5

tests/integrations/litellm/test_litellm.py

@@ -754,6 +754,8 @@ def test_litellm_message_truncation(sentry_init, capture_events):
     assert isinstance(parsed_messages, list)
     assert len(parsed_messages) == 1
     assert "small message 5" in str(parsed_messages[0])
+
+    assert chat_span["data"][SPANDATA.META_GEN_AI_ORIGINAL_INPUT_MESSAGES_LENGTH] == 5
     assert tx["_meta"]["spans"]["0"]["data"]["gen_ai.request.messages"][""]["len"] == 5
 
 

tests/integrations/openai/test_openai.py

@@ -1607,6 +1607,7 @@ def test_ai_client_span_responses_api(
         "gen_ai.usage.total_tokens": 30,
         "gen_ai.request.model": "gpt-4o",
         "gen_ai.response.text": "the model response",
+        "sentry.sdk_meta.gen_ai.input.messages.original_length": 1,
         "thread.id": mock.ANY,
         "thread.name": mock.ANY,
     }
@@ -1910,6 +1911,7 @@ async def test_ai_client_span_responses_async_api(
         "gen_ai.usage.output_tokens.reasoning": 8,
         "gen_ai.usage.total_tokens": 30,
         "gen_ai.response.text": "the model response",
+        "sentry.sdk_meta.gen_ai.input.messages.original_length": 1,
         "thread.id": mock.ANY,
         "thread.name": mock.ANY,
     }
@@ -2177,6 +2179,7 @@ async def test_ai_client_span_streaming_responses_async_api(
         "gen_ai.usage.total_tokens": 30,
         "gen_ai.request.model": "gpt-4o",
         "gen_ai.response.text": "the model response",
+        "sentry.sdk_meta.gen_ai.input.messages.original_length": 1,
         "thread.id": mock.ANY,
         "thread.name": mock.ANY,
     }