`@tanstack/ai-gemini` Error 400 when sending a functionResponse

[hugoattal]

TanStack AI version

@tanstack/ai v0.10.1;

Framework/Library version

@tanstack/ai-gemini v0.8.5

Describe the bug and the steps to reproduce it

I provided a simple tool and try to use it with gemini-3.1-flash-lite-preview, but it always fires a 400 error. The call works, it crashed after the response from my server.

⚠️ EDIT: It seems to work with gemini-2.5-flash, maybe it's on Google side 🤔. Still the duplicate function response seems odd.

I feel like it may be because there are two parts :

• one with text
• one with functionResponse

{
  "parts": [
    {
      "text": "{\"id\":\"root\",\"name\":\"Root\",\"type\":\"folder\",\"children\":[{\"id\":\"dd2f5a5f\",\"name\":\"New page\",\"type\":\"component\"},{\"id\":\"cbadb546\",\"name\":\"New Store\",\"type\":\"store\"}]}"
    },
    {
      "functionResponse": {
        "id": "rmny7t6v",
        "name": "getProjectFileTree",
        "response": {
          "content": "{\"id\":\"root\",\"name\":\"Root\",\"type\":\"folder\",\"children\":[{\"id\":\"dd2f5a5f\",\"name\":\"New page\",\"type\":\"component\"},{\"id\":\"cbadb546\",\"name\":\"New Store\",\"type\":\"store\"}]}"
        }
      }
    }
  ],
  "role": "user"
}

I suspect it may be because of this: https://discuss.ai.google.dev/t/error-400-when-llm-response-contains-both-text-and-functioncall/70826

Also, having the response of the function call in both parts doesn't seem to be very efficient to me... I think the part with text shouldn't be here.

Your Minimal, Reproducible Example - (Sandbox Highly Recommended)

Here's the full payload

{
  "contents": [
    {
      "parts": [
        {
          "text": "Check my project tree and suggest new names"
        }
      ],
      "role": "user"
    },
    {
      "parts": [
        {
          "functionCall": {
            "id": "rmny7t6v",
            "args": {},
            "name": "getProjectFileTree"
          }
        }
      ],
      "role": "model"
    },
    {
      "parts": [
        {
          "text": "{\"id\":\"root\",\"name\":\"Root\",\"type\":\"folder\",\"children\":[{\"id\":\"dd2f5a5f\",\"name\":\"New page\",\"type\":\"component\"},{\"id\":\"cbadb546\",\"name\":\"New Store\",\"type\":\"store\"}]}"
        },
        {
          "functionResponse": {
            "id": "rmny7t6v",
            "name": "getProjectFileTree",
            "response": {
              "content": "{\"id\":\"root\",\"name\":\"Root\",\"type\":\"folder\",\"children\":[{\"id\":\"dd2f5a5f\",\"name\":\"New page\",\"type\":\"component\"},{\"id\":\"cbadb546\",\"name\":\"New Store\",\"type\":\"store\"}]}"
            }
          }
        }
      ],
      "role": "user"
    }
  ],
  "systemInstruction": {
    "parts": [
      {
        "text": "You are **Sidekick**, an expert AI assistant for the **Luna Park** visual editor. [...]\r\n"
      }
    ],
    "role": "user"
  },
  "tools": [
    {
      "functionDeclarations": [
        {
          "name": "getProjectFileTree",
          "description": "Get the project file tree",
          "parameters": {
            "type": "OBJECT",
            "properties": {}
          }
        }
      ]
    }
  ],
  "generationConfig": {}
}

Screenshots or Videos (Optional)

No response

Do you intend to try to help solve this bug with your own PR?

Maybe, I'll investigate and start debugging

Terms & Code of Conduct

• I agree to follow this project's Code of Conduct
• I understand that if my bug cannot be reliable reproduced in a debuggable environment, it will probably not be fixed and this issue may even be closed.

[hugoattal]

Minimum reproduction :

import { chat, toolDefinition } from "@tanstack/ai";
import { createGeminiChat } from "@tanstack/ai-gemini";
import { useChat } from "@tanstack/ai-vue";

const adapter = createGeminiChat("gemma-4-31b-it", "API_KEY");

const coolTool = toolDefinition({
    name: "getCoolStuff",
    description: "Returns cool stuff.",
    inputSchema: {type: "object", properties: {}},
    outputSchema: {type: "object", properties: {stuff: {type: "array", items: {type: "string"}}}},
}).server(() => {
    return ["trains", "rockets"]
});

const aiChat = useChat({
    connection: {
        connect: (messages) => chat({
            adapter,
            messages,
            systemPrompts: ["You're a helpful assistant."],
            tools: [coolTool]
        })
    }
});

await aiChat.sendMessage("List all the cool stuff");

Sends

{
    "parts": [
        {
            "text": "[\"trains\",\"rockets\"]"
        },
        {
            "functionResponse": {
                "id": "6gimt3fv",
                "name": "getCoolStuff",
                "response": {
                    "content": "[\"trains\",\"rockets\"]"
                }
            }
        }
    ]
}