Repository metrics
- Stars
- (4,824 stars)
- PR merge metrics
- (PR metrics pending)
Description
Required Pre-requisites
- I have read the Documentation
- I have searched the Issue Tracker and Discussions that this hasn't been reported yet.
- Consider asking in Discussions first
Motivation
The custom instructions feature allows users to define in natural language what actions are allowed/prohibited for LLM calling a function on our platform.
For example, an user can create a "Work Assistant" agent that can use the app "Gmail". They can specify in the agent custom instruction for Gmail that this agent can only read any work related emails (e.g. by specifying a list of allowed email senders or a list of disallowed ones).
This feature allows more granular level of access control than the one provided by OAuth apps. For example, when you allow the scope of read email for Gmail, then the access token can read any email without restriction. With custom instructions, you can specify more granular levels of access control like the example I mentioned above.
In this issue, we want to add more logging for this feature so we can analyze and further enhance this feature.
Proposed Solution
Add more logging around the custom instructions feature. Please discuss in the issue first before implementing it.