How do I correctly define JSON schema types in my Gemini agent function to avoid a 400 error?

Below is an explanation of the issue and how to resolve it.

---

What the Error Means

The error message:

[GoogleGenerativeAI Error]: Error fetching from <...streamGenerateContent?alt=sse:> [400 Bad Request] * GenerateContentRequest.tools[0].function_declarations[0].parameters.properties[output].type: must be specified when not using one_of

indicates that when generating the structured JSON output, the request includes a tools configuration where the function declaration does not specify a type for the “output” parameter in its JSON schema. In other words, the Gemini API expects the schema to explicitly define the type (for example, as a string, object, array, etc.) for the “output” property when you are not using the one_of construct.

---

How This Relates to Your Workflow

In your automation process, the second agent is expected to extract information and generate a JSON object using a Gemini-based model. The prompt you provide to the agent includes an example JSON schema. However, if the configuration in your node (or prompt) does not declare an explicit type for the “output” property in its function declaration, the Gemini API will respond with a 400 error.

Here’s what you might have in your function declaration (simplified for illustration):

JSONCopied
{
  "function_declarations": [
    {
      "name": "organizeInformation",
      "parameters": {
        "type": "object",
        "properties": {
          "output": {
            // Missing the "type" field here
          }
        }
      }
    }
  ]
}

To fix the error, you need to update the schema so that every property, including “output”, specifies its type. For example:

JSONCopied
{
  "function_declarations": [
    {
      "name": "organizeInformation",
      "parameters": {
        "type": "object",
        "properties": {
          "output": {
            "type": "array",
            "items": {
              "type": "object",
              "properties": {
                "Company name": { "type": "string" },
                "Location": { "type": "string" },
                "Solution name": { "type": "string" },
                "Solution description": { "type": "string" },
                "Solution features": { "type": "array" },
                "Where was it tested?": { "type": "string" },
                "Main clients (oil & gas companies)": { "type": "array" },
                "Economic impact": { "type": "string" },
                "Efficiency": { "type": "string" },
                "Emissions reduction": { "type": "string" },
                "Link to the solution": { "type": "string" },
                "Dimension": { "type": "string" },
                "Discipline": { "type": "string" },
                "Technology status": { "type": "string" }
              },
              "required": [ "Company name", "Location", "Solution name" ]
            }
          }
        }
      }
    }
  ]
}

In this revised schema:

• The output property is explicitly defined as an array of objects.
• Each property inside that object has its type declared.
• This should satisfy the Gemini API’s expectations when not using the one_of schema construct.

---

Recommended Steps

1. Update Your Schema:
   Make sure the function declaration provided to the Gemini model (either within your prompt or in the node's settings) declares the type for every property, including the one causing the error (i.e., output).

2. Test the Schema Locally:
   Before running the full workflow, test the JSON schema configuration by sending a sample request to ensure that the Gemini API accepts it.

3. Adjust Node Parameters if Needed:
   Verify that your n8n node configuration accurately reflects the updated schema so that the JSON output is properly structured for uploading to Google Sheets.

Once you update the schema with the required type information for the "output" property, the Gemini API should no longer return a 400 error, and the second step in your workflow should generate the proper JSON format.

---

This resolution aligns directly with the error message and explains why specifying the type for the "output" parameter is mandatory in your function declaration.