You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In langwatch when having a complex flow with multiple agents and components, the input of the trace shows the input of one of the tools and not the user input
also the LLM traces don't contain usage metrics
Reproduction
deploy a local langflow, add the LANGWATCH_API_KEY to the .env file, and run the financial assistant template
the input of the traces shown is the input of one of the tools used by the agent instead of the user input
Expected behavior
input in langflow trace should match with the user chat input, and llm traces are expected to shown usage metrics and metadata returned by the models
Who can help?
No response
Operating System
mac sequoia 15.2
Langflow Version
1.1.5
Python Version
3.10
Screenshot
Flow File
No response
The text was updated successfully, but these errors were encountered:
The issue you're experiencing with Langflow, where the input trace shows the input of one of the tools instead of the user input, might be related to how Langflow determines which input to display in the trace. Langflow uses the flowPool and flowPoolId to track the flow state and determine the specific output to display. This logic ensures that the correct output is displayed based on the current state of the flow and the specific output name [1].
Regarding the absence of usage metrics in LLM traces, Langflow is supposed to capture usage metrics and metadata through the TracingService class, which utilizes the LangWatchTracer. This includes recording inputs, outputs, and metadata for each trace [2].
To address these issues, you might want to verify the configuration and ensure that the LANGWATCH_API_KEY is correctly set up in your environment. Additionally, reviewing the flow setup to ensure that the user input is correctly mapped and traced might help resolve the input mismatch issue. If the problem persists, it could be a bug in the current version (1.1.5) that might need further investigation or a potential update from the maintainers.
Bug Description
In langwatch when having a complex flow with multiple agents and components, the input of the trace shows the input of one of the tools and not the user input
also the LLM traces don't contain usage metrics
Reproduction
deploy a local langflow, add the LANGWATCH_API_KEY to the .env file, and run the financial assistant template
the input of the traces shown is the input of one of the tools used by the agent instead of the user input
Expected behavior
input in langflow trace should match with the user chat input, and llm traces are expected to shown usage metrics and metadata returned by the models
Who can help?
No response
Operating System
mac sequoia 15.2
Langflow Version
1.1.5
Python Version
3.10
Screenshot
Flow File
No response
The text was updated successfully, but these errors were encountered: