[GEMINI] Field 'include_think' could not be found in request message.

[NLogDEV]

This error occurrs when I use any gemini models for chat/thinking.

400 [{
  "error": {
    "code": 400,
    "message": "Invalid JSON payload received. Unknown name "include_think": Cannot bind query parameter. Field 'include_think' could not be found in request message.",
    "status": "INVALID_ARGUMENT",
    "details": [
      {
        "@type": "type.googleapis.com/google.rpc.BadRequest",
        "fieldViolations": [
          {
            "description": "Invalid JSON payload received. Unknown name "include_think": Cannot bind query parameter. Field 'include_think' could not be found in request message."
          }
        ]
      }
    ]
  }
}
]

[gohoski]

@NLogDEV Hello, sorry for late answering. From what I understood you're trying to use the official Gemini API. Unfortunately I couldn't test out the Gemini API as my VPN IP addresses/Google accounts are randomly getting detected as Russia by AI Studio and AI Studio's API quota is very unreliable, and I added it for convenience in case someone wants to try it out (and I did everything by the Gemini docs correctly). I recommend using Ollama (supports Gemini thinking only) or NavyAI (pretty limited) for Gemini. Gemini is also under heavy limits on free APIs so I recommend you try other models. Right now I'm busy building my new YouTube client so I can't roll out a fix fast, but if we figure out the core issue, the fix will definitely be in v0.3.0.

If you're not using the Gemini API but rather VoidAI, then VoidAI has recently disabled Gemini 2.5/3 for free tiers and you need to use one of the APIs I mentioned above.

What's interesting is that this include_think variable isn't in the Gemini API docs at all for the OpenAI format. https://ai.google.dev/gemini-api/docs/openai It's hard for me to figure out the core issue without being able to communicate with the official API itself.

[NLogDEV]

I think you can modify this to remove the include_think query and see what happens.

also, the gemini docs has thought summaries (scroll down a little bit). You can try it out.

[gohoski]

@NLogDEV Oh that's totally understandable. I didn't even realize that the error could be about URL query parameters. I think I put it in the URL while testing out Upstage but at the end I didn't even add its base URL (due to their models being not very usable). I will try out thought summaries if I will be able to access the Gemini API.

I've changed the URL and built the app as a debug APK (you will have to remove the previous installation of the app beforehand).

numAi-0.2.0-debug.zip

[NLogDEV]

Thanks, it works now!
There's still a problem with it tho (in my testing) that is image attachments dont work. (the model just doesn't respond with anything when attaching images.)

[gohoski]

@NLogDEV That is something that I might not be able to fix at all. numAi converts the images to a Base64 URL and this actually works for all of the OpenAI APIs, but I guess it doesn't work for Gemini's official API. You can try to experiment yourself with the API. Image attachments definitely work for Gemini on Ollama's cloud for me.

I implemented it thru Base64 because the only other option is to provide an image URL, and I don't know of any image hosting that provides proper access via HTTP or TLS 1.0—and besides, the user will need another API key which is just unnecessary.