Agent/Tools - stopping/pausing agent for user input

[whydna]

I'm implementing an agent with toolChoice: 'required' and maxSteps: 100. It may decide, after some steps, that it needs to prompt the user for more information. Once user provides the input, it will be sent back to the agent to continue execution.

From the docs, it seems like the suggested way to stop the agent loop is to have a tool e.g: prompt_for_input with no execute function.

Is the idea that I need to manually check the lastStep/finishReason for this call, and then manually generate the toolResponse for this after prompting the user? Or is there a better way to do this?

[yych42]

I think the idea is that you don't provide execute on the server-side code, so you can use onToolCall on the client side to handle those tool calls. See more here: https://sdk.vercel.ai/docs/ai-sdk-ui/chatbot-with-tool-calling

Did I understand your question right?

[whydna]

I think so. Basically, need to manually construct the ToolResponse after executing off server?

[yych42]

I think so. Basically, need to manually construct the ToolResponse after executing off server?

You could just return the results. Handling probably depends on your framework. Maybe this and this could help?