Configure LongCat-2.5 in Chatbox

One-sentence takeaway

Add an OpenAI API Compatible provider in Chatbox, enter the LongCat API endpoint and key, then add the LongCat-2.5 model to use it.

Use cases

• Suitable tasks: Chat with LongCat-2.5 through Chatbox.

• Unsuitable tasks: Assessing the model's performance based on this guide or inferring configuration steps for other clients.

• Applicable model version: LongCat-2.5.

• Applicable client, agent, or API: Chatbox, using the OpenAI-compatible API.

• Recommended reasoning level and parameters: Not specified.

Ready-to-use content

Test or workflow steps

1. Get an API Key from the LongCat platform and make sure your account has enough token quota.

2. Install and open Chatbox, then select Settings → Model Provider in the lower-left corner.

3. Click Add Provider, select OpenAI API Compatible, and enter the API Host from the table above and your own API Key.

4. Click Check to verify the configuration, then add the model ID LongCat-2.5.

5. Start a new conversation, switch to LongCat-2.5 in the model selector above the input box, and send a test message. Receiving a model response indicates that the configuration succeeded.

Original evidence and data

The official documentation lists LongCat-2.5 as an available model, lists OpenAI/Anthropic API formats, and explicitly provides the API Host and model ID above in its Chatbox configuration steps. The documentation provides no evaluation data.

Scope and limitations

• The documentation says Chatbox supports Windows, macOS, Linux, and mobile devices, and that the mobile setup steps are the same; it does not specify a Chatbox version.

• This is official configuration guidance. It provides no performance tests, inference parameters, or independent verification results.

• This record does not include a real API Key.

Reproduction notes

Follow the configuration steps above. You need a valid LongCat API Key and sufficient quota. Verify the connection by checking whether a new conversation receives a response. This record only checked the official documentation; it did not use an account or make an actual model request.