callawaycloud / llm-apex-agents Goto Github PK

Hi, I am getting errors while deploying the repo to the developer org. Can you please list the steps to deploy?

I have tried deploying the whole main folder from VS Code Windows environment and got the error
"force-app\main\observability\appMenus\AppSwitcher.appMenu-meta.xml In field: AppMenuItem.Name - no ConnectedApp named CPQIntegrationUserApp found"
Moreover, when I deploy using the sfdx force:mdapi:deploy --deploydir main --wait 1000 I get "package.xml not found error". Please help

Currently, if OpenAI returns a 429, the agent exits can cannot be restarted. I might be worth handling this failure by retying the request with a backoff.

Implementing #4 would reduce this issue as the agent could just be restarted.

When I prompt the agent to create a record of a standard or custom object, the action is performed, however the agent does it in a loop until the Queue able job aborts or till maximum heap size is reached.

For example,

Note that in this chain, the record was created thrice before timing out. Am I supposed to prompt it better or is anyone else observing this issue?

I am playing with this incredible piece of art that you have built and struggle with the basics: :-( Where to put the API KEY

Being able to suspend and reanimate agents would really open up the possibility for a lot.

Beyond just recovering from failures, this would be required for "Action Approval" (#6), "User Feedback", "async actions" (such as asking the user for specific input).

Unfortunately, the use of interface is going to make this more challenging as non-concrete types cannot be deserialized...

Should also consider if we can suspend a whole "Stack" of agents to support multi-agent use cases.

For production use, you probably would want some way of approving agent actions before they are executed.

When setting up the tools, you could specify which actions the agent can run without approval.

Would need to modify the current tool registration a bit:

void registerTool(String cmd, IAgentTool tool, Boolean requireApproval)

This break also could also be a chance to provide user feedback and to modify the action the agent is requesting.