Hey there! One quick question if i have an agent with multiple function tools, how can i let it take the output of one function and use it as the input of the subsequent?
Exactly, the controllable part is the thing. I would like to avoid defining a specific tool execution order but also allowing the agent to decide which tool executing based on available data objects (output from other tools)
Is agent’s tool selection only performed through it’s provided context right? What about allowing it to decide right tool using a vector db for instance? It could be an amazing feature
Vector similarity probably is far less reliable than letting the LLM decide. You'd need to have both a rich representation of the tool and a meaningful query to select
Thinking the same! So, in you opinion, what about a ContextRetrievereAgent who is able to know which tools are available for the currently obtained output objects?