What it does
It understood the context of scripts used in middleware operations , execute them in the right sequence and also customize the flow of execution based on user prompt. Also it as gaurdrails support for what to run and what not.
How we built it
It is build on the compute instance, of course google-adk. Specific to adk , below functionalities are used
- Callback as gaurdrail prevent execution of unauthorized commands.
- Multiagent, one for analyzing and one for generating the command is used for seamless execution.
- Iam role binded to computes instance service account is a bare minimum role to do actual task, considering it for security.
Challenges we ran into
In middleware, there are scripts to do task such as healthcheck, startserver, stopserver etc,. For maintenance activity, these scripts need to be executed then and there based on needs, it won't be same in pattern eg: we may need to depend on other server status before taking action on the current server. Writing an orchestrator script is doable, but complex in nature considering the pre-requisites to proceed before each steps. AI agent will remove this complexity by taking user prompt as input and serilaize the script execution according to it.
What's next for Middleware Support Agent
Exploring it to move for a wider use case, which support approval medium and integrating with ticketing tool for auto invocation
Built With
- adk
- gcp
- oci
- python
Log in or sign up for Devpost to join the conversation.