tomsun28 commented on PR #3922:
URL: https://github.com/apache/hertzbeat/pull/3922#issuecomment-3686655132

   > > Hi, i think whether it is provided to the built-in HertzBeat AI or 
exposed as an MCP tool for external agents, the tool’s capabilities should be 
atomic, with the AI model deciding how to orchestrate and invoke them. The 
design should not be driven by traditional user UI interactions, but by 
agent-centric thinking. The required data may be collected in multiple ways—for 
example, the AI may gather it from the user’s prompt, or an agent may read it 
in bulk from text and then add it.
   > 
   > I agree with your point that tools provided for AI calls should have 
higher versatility. I think we could offer some options to users, when they use 
ai assistant in hertzbeat, allowing them to choose the mode of AI involvement. 
For now, let’s refer to these as Protected Mode and Trusted Mode.
   > 
   > In Protected Mode, we could optimize the usage methods and data security 
based on general-purpose tools, enabling AI to integrate deeply into Hertzbeat.
   > 
   > In Trusted Mode, the process is entirely handed over to AI, allowing it to 
fully utilize the existing general-purpose tools to operate Hertzbeat.
   
   HI, the two modes seem like a good approach, would this correspond to two 
sets of Function Tools? Our current AI features are rapidly evolving, so it is 
recommended to use a single set of tools to achieve the desired consistency. 
   Regarding security, the secret data will be transmitted to the LLM for input 
parameter filling, shouldn't we also pay attention to this, considering that 
other data is self-hosted?


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to