Got this error while chatting with an assistant on Dust.tt
ERROR: The assistant attempted to use too many tools. This model error can be safely retried. (code: tool_use_limit_reached)
The assistant had only used one tool - Websearch. So not sure why the error is saying “too many tools”.
My use case needs a collecting top 100 search engine results. So it’s possible that the assistant used search tool too many times and got rated limited by Dust?
How can this use case be achieved without hitting this error?
Hi,
Thanks for flagging this. Have you tried running the assistant again? LLMs can be a bit unpredictable due to their probabilistic nature, so it’s worth giving it another shot.
If you’re still hitting the same error after a couple tries, could you share 2 to 3 conversations URLs? That’d help our engineering team dig into what’s going on.
This doesn’t look like a LLM error. This is something at your end.
Here is a conversation: Dust - Build custom AI assistants to speed up your work
Hey!
Your assistant is a custom assistant based on an anthropic model:
It is tough to fully prevent Anthropic models from using tools when they exhaust their allowed tools. Occasionally, it will do this.
We recommend:
- retry the assistant;
- increase “max tools per run” in the advanced setting (will significantly reduce the likelihood of this happening);
- prompt your assistant to let it know that it may only use his tools n times before replying;
- start a new conversation if none of the solutions above work.
If this behavior is not OK for your use case, we suggest switching to another model until anthropic fixes this specificity.
1 Like
