The community member is a new user of Llamaindex and wants to build an agent using Llama 3.1 that can utilize multiple task-relevant tools simultaneously within a single reasoning step, then consolidate the gathered information to respond. They have implemented a ReAct agent, but it can only use one tool at a time during a search.
In the comments, another community member suggests trying different types of reasoning loops, such as the ones mentioned in the provided links. They also mention the possibility of using Ollama + FunctionCallingAgent, but are unsure if it supports multiple tool calls.
The original community member thanks the suggestion and mentions that they have come across a parameter called allow_parallel_tool_calls, but found that it still only uses one tool at a time within a single reasoning step. They ask under what circumstances this parameter can be used.
The other community members respond that the allow_parallel_tool_calls parameter can be used any time, but it's up to the language model to generate multiple tool calls, which can be hard to control and can't really be "forced".
There is no explicitly marked answer in the comments.
I want to build an agent using Llama 3.1 that can utilize multiple task-relevant tools simultaneously within a single reasoning step, then consolidate the gathered information to respond. How can this be achieved? Currently, I have implemented a ReAct agent, but it can only use one tool at a time during a search.
you could also try Ollama + FunctionCallingAgent. Ollama built function calling support into their API, but not sure if it supports multiple tool calls or not
Thank you for your suggestion, I will give it a try. I also came across a parameter called allow_parallel_tool_calls, but after testing, I found that it still only uses one tool at a time within a single reasoning step.
Do you know under what circumstances this parameter can be used?
