Greasy Fork

The longcat API needs an update, and it would be awesome if you could add Nvidia NIM API support. It would also be great if you could add an option to pin the right-hand menu and keep it as a table of contents for the chats.

Unfortunately, LongCat no longer has a free plan; I will be removing it in the next update. Using NVIDIA NIM as a substitute is a great option! Thanks for the suggestions!

Nice update.

New function to keep the side menu open has been added and will be available in the next update!

Consider a Donation 👉 Ko-fi

Thanks! By the way, I’ve noticed that some models through NVIDIA NIM aren’t working properly, such as Kimi 2.6 and a few others I can’t remember right now.

Some models are also unusually slow, even relatively small ones. I use the same models with Hermes Agent, with instructions that are actually more demanding, and they respond much faster there. I’ve noticed this with models like GLM and Llama 70B, for example.

Could this be related to the way the JavaScript is making the requests, or is there anything that could potentially be optimized?

Also, some models don’t seem to follow even the default prompt. Instead, they behave as if the input were a question addressed directly to them, returning a normal chat response rather than an optimized prompt as instructed.

It would be worth looking into that. As for the pinned chat history, a nice improvement would be to make pinned chats adapt to the website’s interface so they feel like a native part of it, if that makes sense.

Instead of displaying them as a floating element, they could be turned into something more like an integrated index or sidebar section that blends naturally with the existing UI.

I forgot to mention this, but DeepSeek V4 Flash would be a great and simple addition to the NVIDIA NIM options. It’s free, and I believe it could be one of the top-tier models for your script.

1. Model Failures

These failures occur entirely on the provider's side (not just with NVIDIA NIM, but across various providers). Since most of these models are offered for free, they tend to be quite unstable—sometimes support for a model is dropped without notice, or the provider experiences temporary outages. There is nothing wrong with the script here.

2. Performance and Latency

The short answer regarding latency is that there isn't much to be done on our side. The script simply sends the request and waits for the response; no heavy processing happens on the user's side. Response speed depends heavily on the model's capabilities, the provider's infrastructure (providers like Groq tend to be faster, for example), and the length of the system prompt and prompt itself. Paid, dedicated models generally offer much better performance in this regard.

3. Non-Compliance with the Default Prompt

Smaller or lower-capability models frequently ignore parts of the system prompt or respond as if it were a regular chat query. This depends heavily on how well-structured the system prompt is. I strongly recommend replacing the default prompt with a custom one. If you're looking for well-crafted templates, I've included two structured system prompts as a bonus in the Interactive Tutorial: Enhance Prompt that were specifically designed to mitigate these common issues with lower-power models.

4. Pinned Chats UI/UX Improvement

I understand the suggestion, but I won't be making this change. I develop and maintain this project entirely on my own while supporting 34 different platforms, making custom native interfaces for each site completely unsustainable. In the early versions, the prompt button was originally designed to blend in as a native element, but it constantly broke, and as the number of supported platforms grew, I had to unify everything into a single floating visual identity.

Keep in mind that free models are constantly changing or being retired across platforms, so I update them whenever possible. You can also add custom models from any provider you prefer by checking out the User Guide: "Enhance Prompt" Feature.