#Hey guys i am having an issue where a
1 messages · Page 1 of 1 (latest)
Hey, you have several options to prevent this
- Make the LLM condition more specific
- Add context requirement to the condition
- Use the "First message" setting - Make sure Node 1 has a first message that's set to be non-interruptible. This ensures the agent delivers the content before listening for user input that could trigger the condition.
- Add a "buffer" in the condition prompt - Only evaluate as true if the user's response is a substantive answer to the agent's previous question. Generic greetings, acknowledgments, or filler words (yes, hello, okay, uh-huh) should NOT trigger this condition.
- What does that mean ?
- Unfortunately doing that will actually prevent the node switch for a user genuinly says "Yes" for a question asked by the agent right ?
@mortal portal ^^
Context requirement for the condition could be - User has responded to the agent's question about [topic] with [specific intent]
I think the issue that you are facing is the timing of when the condition is evaluated:
Currently: Condition evaluates on any user audio, including the pickup greeting
Desired: Condition should only evaluate after the agent has delivered its content ?
YESS thats right!
Along with this am facing issue with nodes trying to call tools that are not even attached to it.
Idk whats the clean way to fix this!
So something like Condition prompt with context requirement
Do NOT match if user is simply greeting or acknowledging the call.
Only match if the response is clearly answering a specific question the agent asked.```
any solutions for this ?
this is most likely LLM hallucination
yeah am using gemini flash 3.0
1 out of 5 calls have this issue
try adding to your system prompt Your available tools may change between conversation stages. Only use tools currently available to you. or Only use the tools explicitly provided to you. Never attempt to call tools that are not in your available tools list.
Will try, thanks!
sure np, let me know if that helps
When using workflows is it advised uses a smart LLM like claude sonnet as main agent and all nodes will use gemini flash 3.0.
This is my setup rn, is the main Agent LLM actually used when using workflows ?
by main Agent i mean the LLM on the "Agent" tab.
if all the nodes are overwriting agents LLM there is no need it to be different