Repository navigation
Replies: 2 comments
|
Hi @iste2, hope you're doing well! Thanks for writing this up so clearly. Your strawman maps really nicely onto both use cases, and splitting steer from follow up feels like exactly the right line to draw. I was poking around to see how close things already are, and it looks like the CLI does a version of this today with
Python seems to have the same pieces: One caveat: What still seems missing next to your proposal is a public queue API, typed signals, a routing result, and the Maintainers, would it make sense to promote |
|
Hi, we have a few upcoming features that will likely cover this use case:
These are already in the pipeline, so please feel free to comment or give your feedback on these issues. At first glance, this request seems like a duplicate of the linked issues and pull requests, so im going to close out this discussion. If you believe that I have missed something, any your additional scope does not make sense to include in the linked issue, then feel free to re-open this discussion, or create a new one! Thanks! |
Uh oh!
There was an error while loading. Please reload this page.
Summary
I'd like a first-class way to send new input to an agent that is already running, with defined delivery semantics. Today the standard
Agentis strictly turn-based, so apps need their own queue and locking layer on top.Use cases
Prior art
sendMessage/queueMessage/sendSignalwith typed signals, configurable behavior for active vs. idle threads (ifActive/ifIdle), and anacceptedresult that reports the routing decision (wake / deliver / persist / discard / blocked). Input is seen at the next model request; running tools are not cancelled. https://mastra.ai/docs/signals/sending-signals.mdStrawman API (names are just placeholders)
All reactions