Multi-agents: experiment or the future?

[danieldekay]

@assafelovic @ElishaKay - What is your current thinking? Is the langgraph setup just an experiment, or actually the direction GPT-R should develop further?

[assafelovic]

Great question @danieldekay . Would love to see how you think about it. Generally it's not production ready, and is a bit more on the experimental side.

[assafelovic]

Not production ready = langgraph setup

[danieldekay]

I keep reading that people think langchain is in general to be avoided.

Have you thought about a different framework for the agent part?

[danieldekay]

https://github.com/langchain-ai/open_deep_research - this is a langgraph researcher with Tavily integration. Doesn't work fully on my machine with Azure OpenAI, but is fast and looks very useful without any changes.

I wonder if we could compare with GPT-Researcher easily?

[TomLucidor]

I have multiple places to store lists of alternative architectures and multi-agent (I think even code dev work can have multi-agents like MetaGPT but enhanced monoagents like OpenHands are still more effective in non-parallelizable work)
ItzCrazyKns/Perplexica#609 (comment) open-webui/open-webui#9321

[BradKML]

Try socratic dialogues or multi-perspective collaboration. Would need some kind of clear structure that mimics schooling or presentation, and maybe some ways for the user to inject more leading questions for further research?