Feature suggestions

"max_tokens": 163840, "temperature": 0.6, "top_p": 1, "top_k": 50,

Can the setting of custom embedding dimensions be added to the embedding model?

While OpenAI-compatible mode is useful, a dedicated Gemini API mode would greatly enhance the experience for users leveraging Gemini models. Firstly, for models like 'gemini-2.5-pro', it's crucial to display the "thinking process" or intermediate steps (e.g., tool calls, function executions, or complex reasoning steps). Currently, this is not visible, which limits understanding of how the model arrived at its response, especially in multi-turn or complex interactions. Showing these steps would provide valuable insights and improve transparency. Secondly, the Gemini API response body typically includes precise token usage data. When using the Gemini API, it's highly requested that the app utilize these returned values for displaying message token consumption. The app's current internal calculation of token usage can be inaccurate, especially after conversation edits or due to other factors. Using the exact token counts from the API response would ensure accuracy and help users better track their usage and costs.

Typical scenario: I have separate computers at work and at home, and I hope the two devices can sync data; synchronizing data between computer and mobile phone;

I love chatbox but one thing that keeps happening is that I'm loosing long prompts in the middle of writing them (either a shortcut that switch chat, or me wanting to check another thread). For me as soon as one write into a text field (even for an non created chat) it should create a temp thread and store the text live to avoid these issues

The uploaded file pasted_text_0.txt still contains no content—the <FILE_CONTENT> tag is empty. Please ensure the file includes the LaTeX content you want me to review, and I’ll be happy to check it! Let me know once it’s updated.

Dialog titles can be a series or group. You can also search by title. In particular, mobile phone users are particularly difficult to use.

Multiple conversational agents are placed in a group, and each agent can be set individually. After the user outputs a sentence, the agent takes turns to output or randomly disrupts the output. Quantitative changes produce qualitative changes, and the latitude of dialogue (such as moving from two to three dimensions) can produce many unexpected effects and gameplay. The possible downside is that token consumption will increase accordingly, but compared to what you get, these aren't worth mentioning.