Agent doesn't process result of tool call

[IainM22]

I'm using ai version 4.3.4 and after the agent finishes calling a tool it does not respond further and is not aware of the tool result. I can see the tool result in the parts object of the message. From what I understand, there should automatically be a second inference call to the model with the tool result so the agent can format a response, but this doesn't seem to happen (maybe I'm wrong here). I've pasted the messages array from useChat after a successful call to the add tool below.

Additionally, after the agent doesn't respond if I ask it what the result of the tool call was, it thinks it hasn't called the tool and tries again. I'm using streamText on my /chat endpoint

[
    {
        "id": "NJSMqGq9JgEaPuCP",
        "createdAt": "2025-04-13T03:35:04.811Z",
        "role": "user",
        "content": "add 2 and 3",
        "parts": [
            {
                "type": "text",
                "text": "add 2 and 3"
            }
        ]
    },
    {
        "id": "msg-wLTVEcEqTXagOjMjRdKUdXes",
        "createdAt": "2025-04-13T03:35:13.673Z",
        "role": "assistant",
        "content": "Certainly! I can help you add 2 and 3 using the \"add\" function available to me. Let me use this tool to calculate the result for you.",
        "parts": [
            {
                "type": "step-start"
            },
            {
                "type": "text",
                "text": "Certainly! I can help you add 2 and 3 using the \"add\" function available to me. Let me use this tool to calculate the result for you."
            },
            {
                "type": "tool-invocation",
                "toolInvocation": {
                    "state": "result",
                    "step": 0,
                    "toolCallId": "toolu_01XNLZTVQNxvjUeumM6PDTb3",
                    "toolName": "add",
                    "args": {
                        "a": 2,
                        "b": 3
                    },
                    "result": {
                        "content": [
                            {
                                "type": "text",
                                "text": "5"
                            }
                        ]
                    }
                }
            }
        ],
        "toolInvocations": [
            {
                "state": "result",
                "step": 0,
                "toolCallId": "toolu_01XNLZTVQNxvjUeumM6PDTb3",
                "toolName": "add",
                "args": {
                    "a": 2,
                    "b": 3
                },
                "result": {
                    "content": [
                        {
                            "type": "text",
                            "text": "5"
                        }
                    ]
                }
            }
        ],
        "revisionId": "8xY6aI9ku6IPhImL"
    }
]

[IainM22]

It turns out this was due to me not setting the maxSteps option in streamText, which defaults to 1. Once I set it to a number greater than 1 the agent processed the result

[ignaciogiri]

Thank you and Google for this. I spent 2 hours debugging a silent error.

[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