AI Agent output after execution is empty i

[takefy-dev]

Bug Description

When an ai call tool sometimes the final output is empty and it cause redis to fail saving the conversation data so the ai just do wrong thing

To Reproduce

Create an ai agent put him a tool with redis store and it will happens when you try to trigger the tool but it does not happens everytim tho

Expected behavior

To output something so it saves the tool execution and continue the flow

Operating System

Linux

n8n Version

1.62.3

Node.js Version

22

Database

PostgreSQL

Execution mode

main (default)

[Joffcom]

Hey @takefy-dev,

We have created an internal ticket to look into this which we will be tracking as "GHC-308"

[Joffcom]

Hey @takefy-dev,

Are you able to share a minimal workflow and data set that reproduces this issue?

[takefy-dev]

create a flow with antropic claude 3.5 put chat input + redis memory + a http tool that makes a request and return response: "a string" make a long prompt and try sometimes it will output this error...

[takefy-dev]

any news it so anoying?

[takefy-dev]

#11300 ok this issue is a tempory fix but the wierd thing is ai not generating an ouput while when i test the exact same msg in the playground it output always a msg. Test 5 time in n8n no output, test 5 time in playground always success. The only difference in n8n it has trigger a tool so may be. So what it will output if output string is empty also it will save??

[takefy-dev]

Yeah i confirm it's fully bug i just used v 1.64.1 and now it save the conversation of the human input but since not output is provided it never save the ai output so we have error like this:
A conversation must alternate between user and assistant roles. Make sure the conversation alternates between user and assistant roles and try again.

Why only in n8n the llm after a tool call is outputing nothing

[takefy-dev]

@Joffcom

[janober]

Fix got released with [email protected]

[takefy-dev]

Fix got released with [email protected]

Now getting this with aws bedrock...
It's because the output of the last llm execution output is empty...

Then when you tried to chat after the issue:
this because in the store the last key is a user role and not assistant...