Hi everyone, I’m currently testing Google’s ADK and building my first multi-agent system (a customer service agent). My idea is to have a root agent that communicates directly with the customer, while the sub-agents only interact with or assist the root agent (not the customer directly). The problem I’m facing is that all agents are responding to the customer, which breaks some intended features. For example: - The root agent can open a ticket using a tool when it cannot solve the customer’s problem. - If the customer’s prompt is about an order and the root agent delegates it to the “orders” sub-agent, the sub-agent should send its answer back to the root agent. - However, what happens now is that the sub-agent replies directly to the customer saying it can’t do anything, instead of sending this information to the root agent, which should then decide whether to answer or propose opening a ticket. Has anyone faced this issue before or knows how to make only the root agent respond to the customer?

Right now I'm designing a pipeline/workflow using CrewAI flows - but the tech stack may not be important. My workflow takes in a list of elements as its input, a list of json objects for example, performs the pipeline/workflow on each input and then returns the complete results. My question is: What is the best practice for when to split the inputs and gather the results. In my immediate example, do I create the flow to act on 1 input and have my runner (fastapi in this case) kick off a new instance of the crew for each input and then gather the results of all the flow instances into 1 result? Or, do I have the workflow (in this case CrewAI flows) have the looping mechanism internal to process and then return which would mean 1 outer flow kickoff with a list as input. I can see there being a "it depends" answer based on complicated state or expensive to create objects, but I feel like this is a pretty common use case, so there's probably some great common wisdom I should draw from.

I am building a multi crew workflow within a single flow and I implemented async function calling to kickoff both crews simultaneously and within one of the crews there is async execution switched on for the end tasks. In few of the runs, the entire flow unexpectedly switches off as there more processing steps after gathering both outputs. I also implemented another approach where I would have multiple start decorators with async initialization and utilize and_ to gather as a sync point and though in some runs I get full working setup flow, sometimes it abruptly stops. Would love to open a detailed discussion share code etc on this and along with also start a discussion on the following topics 1) parallel flow executions 2) flow nesting 3) parallel tool execution 4) Hierarchical processes 5) intelligent way of ensuring agent takes appropriate input and passes to custom tool rather than specifying tool input Would love to get feedback, open a discussion and learn more from the community in areas where I still have to learn /if I am wrong or have not particularly understood a concept and it's implementation. I love testing out new ways to frameworks.

I have lots of structure json files with title and description (could be job listing, property listing etc.). I'm trying to get a crew to return the top matching items for a given query title. All the examples I've seen work on a small example set but fail to handle large collections of documents like above. I use local LLM and often get warnings that the input prompt token count exceeded the limit. Tried the pro tools approach vs. the noob, but still struggles. A working example would be nice.

Brandon, I am trying to run (409) CrewAI RAG Deep Dive [Basic & Advanced Examples] - YouTube this code on my own, and getting errors due. I am trying to figure it out. I need this example for a project I am working internally. Thanks for your help. Because no versions of crewai match >0.118.0,<0.119.0 and crewai (0.118.0) depends on chromadb (>=0.5.23), crewai (>=0.118.0,<0.119.0) requires chromadb (>=0.5.23). And because crewai-tools (0.4.26) depends on chromadb (>=0.4.22,<0.5.0) and no versions of crewai-tools match >0.4.26,<0.5.0, crewai (>=0.118.0,<0.119.0) is incompatible with crewai-tools (>=0.4.26,<0.5.0). So, because crewai-rag-basics depends on both crewai-tools (^0.4.26) and crewai (^0.118.0), version solving failed.