How can I get more input from the user after an agent is triggered?

[pnewhook]

Is there an example or guidance on getting more input from a user once an agent has been selected? I'm creating a restaurant recommender that recommends a restaurant based on the preferences and location of the user. It's possible the user won't specify their location. How can I go back to the user to ask their location, then resume the agent flow? I've embedded embabel in a web service, so I can't use any shell functionality. I think this example relies on shell execution, https://github.com/embabel/embabel-agent-examples/blob/main/examples-java/src/main/java/com/embabel/example/handoff/Handoff.java.

[igordayen]

Incidently im also working on Restaurant Planner as meta-agent (private repo) as sub-project. Meta-agent can generate any Agent, so i chose to experiment with restaurant planner. Started with Shell-based meta-agent. would be happy to collaborate. please reach me out on discord @igordayen thank you

[igordayen]

you may want to check Tripper repo, its web-based sofisticated Travel Planner acting as Embabel Agent. Thanks.

[johnsonr]

See the StarNewsFinder in embabel-agent-examples and what happens if a StarPerson cannot be created from user input.

[pnewhook]

Thanks @johnsonr. My understanding is fromForm() throws an exception under the hood. The shell start knows how to handle that exception and collect input from the user. I'm currently hosting my agents in a Spring Boot web app, and triggering them with Autonomy.chooseAndRunAgent().

1. Is that the correct way to trigger agents if I want embabel to select the appropriate agent based on user input.
2. Is there any built in support or planned support for that kind of error handling when agents are hosted in a web app?
3. If I host my agents in an MCP server would there be a way to use Elicitation to collect user input.

[schabiyo-eng]

I was just looking for the same @pnewhook let me know if you got an answer

[LiveNathan]

I have all of the same questions. For number 1, I'm currently handling it like this, but your solution sounds better.

UserInput userInput = new UserInput(aiRequest.messageFromUser());
AgentInvocation<String> invocation = AgentInvocation.create(agentPlatform, String.class);
String response = invocation.invoke(userInput);

[igordayen]

hi @pnewhook - curious on business side which business APIs you are using: opentable, yelp, resy, etc?