Explore Self Learning Capabilities

[chilladelia]

I receive many requests from product managers to explore self learning capabilities of agents.

In my opinion it is a very interesting field.

Currently Neuron supports chat histories and has RAG capabilities. But there are other long term memory approaches out there, such as mem0 and zep.

Would be great to explore the possibilities with Neuron. For instance, by providing tools for mem0, etc but maybe also exploring native possibilities without additional tools.

The idea is to have generic agents that learn across different user sessions.

[ilvalerione]

Hi @chilladelia , it still a very interesting topic. For what can be done with external services, I already explored a bit this environment using Zep knowledge graph, and I implemented a tool for it: https://docs.neuron-ai.dev/the-basics/tools#zep-memory

You could start from there and eventually share here some findings of idea to improve in this direction.

Using well structured tools it could be go quite far also storing fact, relevant informations, etc, directly summarized from the LLM, and maintain this information on files, database, whetever.

[chilladelia]

Wow cool, missed the Zep example! Will test that.

[chilladelia]

First findings:

• ZepSearchGraphTool fails, if there are no results in the store (created PR Avoid exception if zep returns empty result. #339).
• the tool doesn't work for me properly yet. For instance, it forgets my name most of the time. Don't know why is that, still need to figure out the right prompting.