Skip to content

fix(langchain): Set agent name as gen_ai.agent.name for chat and tool spans #5877

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
alexander-alderman-webb wants to merge 9 commits into
base: master
Choose a base branch
from
+54 −92
Open

fix(langchain): Set agent name as gen_ai.agent.name for chat and tool spans #5877

Show file tree
Hide file tree
Changes from 5 commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
d4c4cd4
fix(langchain): Set agent name as gen_ai.agent.name
alexander-alderman-webb Mar 25, 2026
181d5cb
merge master
alexander-alderman-webb Mar 31, 2026
8cf3f81
typing
alexander-alderman-webb Mar 31, 2026
56ec48f
fix span description
alexander-alderman-webb Mar 31, 2026
d8c06f8
defensive check
alexander-alderman-webb Mar 31, 2026
36ca817
no agent name in stream
alexander-alderman-webb Mar 31, 2026
0d43616
simplify
alexander-alderman-webb Mar 31, 2026
b9387b8
set run name
alexander-alderman-webb Apr 1, 2026
cf70d07
Merge branch 'master' into webb/langchain/agent-name
alexander-alderman-webb Apr 1, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
162 changes: 66 additions & 96 deletions sentry_sdk/integrations/langchain.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,3 @@
import contextvars
import itertools
import sys
import json
Expand Down Expand Up @@ -162,44 +161,6 @@
return content


# Contextvar to track agent names in a stack for re-entrant agent support
_agent_stack: "contextvars.ContextVar[Optional[List[Optional[str]]]]" = (
contextvars.ContextVar("langchain_agent_stack", default=None)
)


def _push_agent(agent_name: "Optional[str]") -> None:
Copy link
Copy Markdown
Member

@ericapisani ericapisani Apr 1, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice to be getting rid of this code 🔥

"""Push an agent name onto the stack."""
stack = _agent_stack.get()
if stack is None:
stack = []
else:
# Copy the list to maintain contextvar isolation across async contexts
stack = stack.copy()
stack.append(agent_name)
_agent_stack.set(stack)


def _pop_agent() -> "Optional[str]":
"""Pop an agent name from the stack and return it."""
stack = _agent_stack.get()
if stack:
# Copy the list to maintain contextvar isolation across async contexts
stack = stack.copy()
agent_name = stack.pop()
_agent_stack.set(stack)
return agent_name
return None


def _get_current_agent() -> "Optional[str]":
"""Get the current agent name (top of stack) without removing it."""
stack = _agent_stack.get()
if stack:
return stack[-1]
return None


def _get_system_instructions(messages: "List[List[BaseMessage]]") -> "List[str]":
system_instructions = []

Expand Down Expand Up @@ -465,9 +426,11 @@
if ai_system:
span.set_data(SPANDATA.GEN_AI_SYSTEM, ai_system)

agent_name = _get_current_agent()
if agent_name:
span.set_data(SPANDATA.GEN_AI_AGENT_NAME, agent_name)
agent_metadata = kwargs.get("metadata")
if isinstance(agent_metadata, dict) and "lc_agent_name" in agent_metadata:
span.set_data(
SPANDATA.GEN_AI_AGENT_NAME, agent_metadata["lc_agent_name"]
)

for key, attribute in DATA_FIELDS.items():
if key in all_params and all_params[key] is not None:
Expand Down Expand Up @@ -665,9 +628,11 @@
if tool_description is not None:
span.set_data(SPANDATA.GEN_AI_TOOL_DESCRIPTION, tool_description)

agent_name = _get_current_agent()
if agent_name:
span.set_data(SPANDATA.GEN_AI_AGENT_NAME, agent_name)
agent_metadata = kwargs.get("metadata")
if isinstance(agent_metadata, dict) and "lc_agent_name" in agent_metadata:
span.set_data(
SPANDATA.GEN_AI_AGENT_NAME, agent_metadata["lc_agent_name"]
)

if should_send_default_pii() and self.include_prompts:
set_data_normalized(
Expand Down Expand Up @@ -793,9 +758,7 @@
span.set_data(SPANDATA.GEN_AI_USAGE_TOTAL_TOKENS, total_tokens)


def _get_request_data(
obj: "Any", args: "Any", kwargs: "Any"
) -> "tuple[Optional[str], Optional[List[Any]]]":
def _get_available_tools(obj: "Any") -> "Optional[List[Any]]":
"""
Get the agent name and available tools for the agent.
"""
Expand All @@ -810,16 +773,23 @@
)
tools = tools if tools and len(tools) > 0 else None

return tools


def _get_run_name(obj: "Any", args: "Any") -> "Optional[str]":
agent = getattr(obj, "agent", None)
runnable = getattr(agent, "runnable", None)
runnable_config = getattr(runnable, "config", {})
try:
agent_name = None
if len(args) > 1:
agent_name = args[1].get("run_name")
if agent_name is None:
agent_name = runnable_config.get("run_name")
except Exception:
pass

return (agent_name, tools)
return agent_name

Check failure on line 792 in sentry_sdk/integrations/langchain.py

View check run for this annotation

@sentry/warden / warden: code-review

UnboundLocalError when exception occurs in _get_run_name 

In `_get_run_name`, the variable `agent_name` is only defined inside the `try` block (line 784). If an exception occurs before the assignment (e.g., `args` doesn't support `len()` or indexing), the `except` block swallows the exception, but the `return agent_name` on line 792 will raise `UnboundLocalError: local variable 'agent_name' referenced before assignment`. This will cause the langchain integration to fail when invoking agents with unexpected argument types.


def _simplify_langchain_tools(tools: "Any") -> "Optional[List[Any]]":
Expand Down Expand Up @@ -987,58 +957,53 @@
if integration is None:
return f(self, *args, **kwargs)

agent_name, tools = _get_request_data(self, args, kwargs)
start_span_function = get_start_span_function()

run_name = _get_run_name(self, args)
with start_span_function(
op=OP.GEN_AI_INVOKE_AGENT,
name=f"invoke_agent {agent_name}" if agent_name else "invoke_agent",
name=f"invoke_agent {run_name}" if run_name else "invoke_agent",
origin=LangchainIntegration.origin,
) as span:
_push_agent(agent_name)
try:
if agent_name:
span.set_data(SPANDATA.GEN_AI_AGENT_NAME, agent_name)
if run_name:
span.set_data(SPANDATA.GEN_AI_AGENT_NAME, run_name)

Check warning on line 968 in sentry_sdk/integrations/langchain.py

View check run for this annotation

@sentry/warden / warden: find-bugs

new_invoke ignores lc_agent_name metadata unlike new_stream 

The `new_invoke` function does not check `kwargs.get('metadata', {}).get('lc_agent_name')` like `new_stream` does. When users pass `metadata={'lc_agent_name': 'MyAgent'}` to `invoke()`, the agent name from metadata is ignored and only `run_name` (from args) is considered. This contradicts the PR's stated goal of setting agent name as `gen_ai.agent.name` and creates inconsistent behavior between `invoke()` and `stream()` methods.

span.set_data(SPANDATA.GEN_AI_OPERATION_NAME, "invoke_agent")
span.set_data(SPANDATA.GEN_AI_RESPONSE_STREAMING, False)
span.set_data(SPANDATA.GEN_AI_OPERATION_NAME, "invoke_agent")
span.set_data(SPANDATA.GEN_AI_RESPONSE_STREAMING, False)

_set_tools_on_span(span, tools)
tools = _get_available_tools(self)
_set_tools_on_span(span, tools)

# Run the agent
result = f(self, *args, **kwargs)
# Run the agent
result = f(self, *args, **kwargs)

input = result.get("input")
if (
input is not None
and should_send_default_pii()
and integration.include_prompts
):
normalized_messages = normalize_message_roles([input])
scope = sentry_sdk.get_current_scope()
messages_data = truncate_and_annotate_messages(
normalized_messages, span, scope
input = result.get("input")
if (
input is not None
and should_send_default_pii()
and integration.include_prompts
):
normalized_messages = normalize_message_roles([input])
scope = sentry_sdk.get_current_scope()
messages_data = truncate_and_annotate_messages(
normalized_messages, span, scope
)
if messages_data is not None:
set_data_normalized(
span,
SPANDATA.GEN_AI_REQUEST_MESSAGES,
messages_data,
unpack=False,
)
if messages_data is not None:
set_data_normalized(
span,
SPANDATA.GEN_AI_REQUEST_MESSAGES,
messages_data,
unpack=False,
)

output = result.get("output")
if (
output is not None
and should_send_default_pii()
and integration.include_prompts
):
set_data_normalized(span, SPANDATA.GEN_AI_RESPONSE_TEXT, output)
output = result.get("output")
if (
output is not None
and should_send_default_pii()
and integration.include_prompts
):
set_data_normalized(span, SPANDATA.GEN_AI_RESPONSE_TEXT, output)

return result
finally:
# Ensure agent is popped even if an exception occurs
_pop_agent()
return result

return new_invoke

Expand All @@ -1050,24 +1015,31 @@
if integration is None:
return f(self, *args, **kwargs)

agent_name, tools = _get_request_data(self, args, kwargs)
start_span_function = get_start_span_function()

agent_name = kwargs.get("metadata", {}).get("lc_agent_name")
run_name = _get_run_name(self, args)

span_name = "invoke_agent"
if agent_name is not None:
span_name = f"invoke_agent {agent_name}"
elif run_name:
span_name = f"invoke_agent {run_name}"

span = start_span_function(
op=OP.GEN_AI_INVOKE_AGENT,
name=f"invoke_agent {agent_name}" if agent_name else "invoke_agent",
name=span_name,
origin=LangchainIntegration.origin,
)
span.__enter__()

_push_agent(agent_name)

if agent_name:
if agent_name is not None:
span.set_data(SPANDATA.GEN_AI_AGENT_NAME, agent_name)

Check warning on line 1037 in sentry_sdk/integrations/langchain.py

View check run for this annotation

@sentry/warden / warden: code-review

GEN_AI_AGENT_NAME not set when using run_name as fallback in new_stream 

In `new_stream`, when `agent_name` is None but `run_name` is available, the span name includes `run_name` (line 1027) but `GEN_AI_AGENT_NAME` is never set (line 1036 only checks `agent_name`). This is inconsistent with `new_invoke` which sets `GEN_AI_AGENT_NAME` whenever `run_name` is available (line 967-968). This causes telemetry data to be incomplete when streaming with a run_name but no explicit agent_name.

span.set_data(SPANDATA.GEN_AI_OPERATION_NAME, "invoke_agent")
span.set_data(SPANDATA.GEN_AI_RESPONSE_STREAMING, True)

tools = _get_available_tools(self)
_set_tools_on_span(span, tools)

input = args[0].get("input") if len(args) >= 1 else None
Expand Down Expand Up @@ -1117,7 +1089,6 @@
raise
finally:
# Ensure cleanup happens even if iterator is abandoned or fails
_pop_agent()
span.__exit__(*exc_info)

async def new_iterator_async() -> "AsyncIterator[Any]":
Expand All @@ -1143,7 +1114,6 @@
raise
finally:
# Ensure cleanup happens even if iterator is abandoned or fails
_pop_agent()
span.__exit__(*exc_info)

if str(type(result)) == "<class 'async_generator'>":
Expand Down
6 changes: 6 additions & 0 deletions tests/integrations/langchain/test_langchain.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -259,6 +259,8 @@ def test_langchain_create_agent(
assert chat_spans[0]["origin"] == "auto.ai.langchain"

assert chat_spans[0]["data"]["gen_ai.system"] == "openai-chat"
assert chat_spans[0]["data"]["gen_ai.agent.name"] == "word_length_agent"

assert chat_spans[0]["data"]["gen_ai.usage.input_tokens"] == 10
assert chat_spans[0]["data"]["gen_ai.usage.output_tokens"] == 20
assert chat_spans[0]["data"]["gen_ai.usage.total_tokens"] == 30
Expand Down Expand Up @@ -415,6 +417,10 @@ def test_tool_execution_span(
assert chat_spans[1]["origin"] == "auto.ai.langchain"
assert tool_exec_span["origin"] == "auto.ai.langchain"

assert chat_spans[0]["data"]["gen_ai.agent.name"] == "word_length_agent"
assert chat_spans[1]["data"]["gen_ai.agent.name"] == "word_length_agent"
assert tool_exec_span["data"]["gen_ai.agent.name"] == "word_length_agent"

assert chat_spans[0]["data"]["gen_ai.usage.input_tokens"] == 142
assert chat_spans[0]["data"]["gen_ai.usage.output_tokens"] == 50
assert chat_spans[0]["data"]["gen_ai.usage.total_tokens"] == 192
Expand Down
Loading