search

XiaoMi / ozhera

Application Observable Platform in the Cloud Native Era
https://ozhera.m.one.mi.com
Apache License 2.0
59 stars 28 forks source link

feat: add intelligence method #380

Open sadadw1 opened 1 week ago

sadadw1 commented 1 week ago

feat: add intelligence method #368

goodjava commented 1 week ago

I think this is a very good idea. Could you specifically tell us how to reduce the cost for people to troubleshoot problems through Hera? Also mention the systems this system depends on and the timing of its future release.

sadadw1 commented 1 week ago

I think this is a very good idea. Could you specifically tell us how to reduce the cost for people to troubleshoot problems through Hera? Also mention the systems this system depends on and the timing of its future release.

I believe that reducing costs is mainly reflected in two aspects:

The main external dependency of this module is the AI large model, and all data given to the AI large model comes from ozhera.

At present, the intelligent functions are just a framework, and the specific implementation will be carried out from time to time, so please stay tuned.

goodjava commented 4 days ago

This project seems to have AI backend dependencies. Have you considered relying on open-source AI backends like Dify in the later stages? By the way, do you have any plans to open-source your AI backend?

sadadw1 commented 4 days ago
  1. Not relying on an AI backend is to make OzHera lighter, reducing the costs for users in terms of building, launching, running, debugging, and maintaining.
  2. Internally at Xiaomi, we rely on a backend similar to Dify, called the M78 BOT platform.
  3. This platform may be open-sourced in the future, and at that time, OzHera will also support the ability to rely on M78. I believe this will make OzHera's capabilities even stronger.