In V2 it appears that the ‘show_tool_calls’ parameter has been removed from the Agent call list (i.e., an error is generated indicating that it is an invalid argument). I may have missed it, but how do you now get the tools that are used when you are debugging a work flow? A ‘reasoning’ LLM may consider using one of several tools, but you may not be able to tell which one if any it chose to use… It might have also had the information cached from a previous call, and uses history instead…
Hi @bills, thanks for reaching out and supporting Agno. I’ve shared this with the team, we’re working through all requests one by one and will get back to you soon.If it’s urgent, please let us know. We appreciate your patience!
We have already found that enabling debug_mode (=True) shows the tool calls. BUT, it also turns on pages of other diagnostic information that is only interesting when you are doing a deep dive into some issue. As you know, a number of LLMs do not handle tool calling uniformly and consistently. Some issues are fixed in the LLM themselves, and others in the intermediate software (frameworks).
When something starts to malfunction and/or becomes non-repeatable, one of the first things we do was to turn on tool_calls and examine the requests and responses. In some cases that helped us modify the instructions/prompts to the particular LLM that fixed the problem. Turning on debug_mode now generates reams of data that we have to wade through… And since we are working with 20+ LLMs, our Q/Aing is taking substantially longer…
So please consider putting it back in. Since we are routing everything through a single entry point, reenabling and using the switch would be extremely helpful.
Were you using .run before in v1 code and it displayed the tool calls? You can try using print_response to see the tool calls made without the results.
Yes, we are migrating our code from V1 to V2. Most of our agents and teams go through a small common set of code into your framework. So changing to use ‘print_response’ from ‘show_tool_calls’ is a simple change. Thanks.
You might consider adding a note to the v1 to v2 migration documentation. We have had all sorts of challenges with inconsistant and unexpected LLM responses and major tweaking between various and the same models. The ‘show_tool_calls’ was invaluable.