I’m wondering if anyone else can spin up the example and see if they have the same issue with the ‘useChat with tools’ example.
I just asked a question like “Who is kenye west” and the response get’s cut off after a couple lines. From testingI can see that the server is sending the full response, so it’s an issue on the client side. I’m not even sure how to debug this.
Hey @randulajariyawanse. If you only run into this issue locally, that sounds like a local network issue. If it’s happening when deployed with Vercel then you may be able to find more info in the runtime logs.
One of the most common reasons for cut-off responses is timeouts. By default, Vercel has a 10 second timeout for serverless functions on the Hobby plan, which may not be enough for longer AI-generated responses. You can increase the maximum duration of your function by adding export const maxDuration = 60; to the function file.
Server side errors are another possibility. These should also appear in the logs. If there’s an error on the server-side during the streaming process, it could cause the stream to end prematurely. Make sure to implement error handling.
Sometimes, network issues can cause streaming responses to be cut off. Make sure there is a stable connection and that there are no firewall or proxy issues blocking the stream.
Keep in mind that resource limitations could also be a factor if the AI model is generating very long responses. Consider breaking up long generations into smaller chunks or implementing pagination if that’s the case.