Replies: 2 comments
-
Right now, the only ways to force the agent to stop are the following (although feel free to correct me if I'm missing methods):
I think for the first method, other people have had this problem as well, as seen on : #10714 Having an endpoint on the agent excutor that can be called at any point which would stop the chain, could help with this issue. I have a solution available at: 5085990 This would require less change (which would make it backward compatible) as it would not need a revamp of how the tools return (for example, changing from strings to objects with flags). It would also allow the chain designer to force terminate at any point they want. I will send an MR soon. |
Beta Was this translation helpful? Give feedback.
-
MR was sent in: #22327 |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Checked
Feature request
Right now return direct will terminate the change upon completion of the underlying tool; this is the case regardless of whether or not the tool threw an exception. I want the ability to have the agent chain not terminate if an exception was thrown, otherwise to terminate.
Motivation
Some tools may handle errors gracefully and raise a custom exception with details on how the agent can fix its mistakes. In those cases we want the agent to retry. When successful though, we want the chain to end. This behaviour is not currently possible, its either all or nothing.
Proposal (If applicable)
No response
Beta Was this translation helpful? Give feedback.
All reactions