i don't understand something about tool calling loop in ai sdk

[th0x11]

Hi everyone 👋

I’m building an agent with AI SDK v5 that has access to tools (like webSearch).

My goal: let the agent run in a truly agentic way — it could make 0, 1, 2, or many tool calls, depending on what’s needed. Basically, I don’t want to hard-set the number of steps, I just want it to run until it decides it’s done.

Questions:

What’s the right way to do this in AI SDK v5?

Specifically, with stepCountIs(n)
: does the n mean exactly n steps, an upper bound, or “run until step n and then stop”?
(before in ai sdk v4 I just choosed maxSteps(10000) and it was okay)

And if I want open-ended behavior (agent free to stop whenever), should I just avoid stepCountIs and loop manually? Or is there a built-in pattern for “run until done but cap at max N to avoid infinite loops”?

Thanks 🙏

[muniter]

You can provide a finished tool that the agent can call to let you know is done, and use stopWhen: hasToolCall('finish')

https://ai-sdk.dev/docs/reference/ai-sdk-core/stream-text#stop-when

[th0x11]

Thank you very much !

[lgrammel]

The agentic loop stops in 4 cases:

• a finish reasoning other than tool-calls is returned
• a tool that is invoked does not have an execute function
• a tool call needs approval
• the condition from stopWhen is met

[th0x11]

Thanks a lot for the explaination!

[github-actions[bot]]

This discussion was automatically locked because it has not been updated in over 30 days. If you still have questions about this topic, please ask us at community.vercel.com/ai-sdk