Skip to main content

​
Before you start

  • Chatbox: download it from chatboxai.app for Windows 10 or later (64-bit), macOS 11 or later, Linux (AppImage or deb), iOS, or Android, or open the web version at web.chatboxai.app.
  • API key: create one in Tokens with the default group and the Usage first billing mode. If you also add a Claude API Compatible provider, create a second key in the Claude Code group for it.
On first launch, Chatbox shows a welcome screen. Since you bring your own key, choose the option that skips the guided setup.

​
Add LaoZhang API

1

Create a custom provider

Open Settings → Model Provider, click Add at the bottom of the provider list, and choose Add Custom Provider. On mobile, tap ☰ at the top left to reach Settings.Name the provider “LaoZhang API” and set the API mode to OpenAI API Compatible. This mode works with every chat model in the catalog, so start with it.
2

Enter the key and host

  • API Key: your LaoZhang API key
  • API Host: https://api.laozhang.ai/v1
  • API Path: leave empty
Check the preview line below API Path. It should read https://api.laozhang.ai/v1/chat/completions, with /v1 appearing once.Leave the network-compatibility option (Improve network compatibility) off. Turn it on only if the connection fails; in the web version, it routes your requests through Chatbox’s servers.
3

Add models

In the model section, click Fetch to load the model list from LaoZhang API and pick the ones you want, or click New to enter a model ID such as claude-sonnet-5-5. Model IDs must match the model catalog exactly.
4

Check the connection

Click Check next to the API key. It needs at least one model in the list. Chatbox also detects whether each model supports vision and tool use; reasoning is not detected, so tick it yourself for thinking models.To test a single model, open it and click Test Model.
Start a new chat and pick a LaoZhang API model from the model selector. If it replies, the setup works.

​
Choose a model

Prices are in the model catalog.

​
Use Claude, Gemini, or Responses formats

The OpenAI API Compatible mode covers ordinary chat. Choose another API mode when you want a model’s native request format, for example to use its thinking controls. Each mode builds the request URL differently, so the API Host value changes with it:
  • Claude API Compatible: the host must include /v1, because Chatbox appends only /messages. Use it for Claude models such as claude-sonnet-5-5; GLM-5.2 and the DeepSeek V4 models also work here. This mode needs a key in the Claude Code token group.
  • Google Gemini API Compatible: the host must not include /v1, because Chatbox adds /v1beta itself. Use it for Gemini models.
  • OpenAI Responses API Compatible: keep the default API Path, /responses. Use it for GPT models.
The OpenAI Responses API Compatible mode needs Chatbox 1.18.1 or later, and Google Gemini API Compatible needs 1.17.0 or later. Fetch works in every mode. The OpenAI, Responses, and Gemini modes share your default-group key, while Claude API Compatible uses the Claude Code-group key. A simple setup is one provider per mode, each holding the models that belong to it.

​
Thinking controls

Chatbox enables the thinking control only when the API mode matches the model family: In any other combination, the thinking control is disabled. Also tick Reasoning in the model’s capabilities, since Check does not detect it.

​
Generate images

Add an image model ID from the model catalog to the OpenAI API Compatible provider, open the model, and set Model Type to Image. Chatbox then sends requests to https://api.laozhang.ai/v1/images/generations on its own. Keep API Host and API Path exactly as they are for chat; don’t change the path for image models. For image models and request options, see the image generation guide.

​
Use MCP tools

MCP tools work only in Work mode in the desktop app.
1

Add an MCP server

Open Settings → MCP and add a server. Remote servers (HTTP or SSE) can be added on every platform; local servers (stdio) only on desktop. Click Test, and save once the test passes.
2

Enable tool use on the model

Open the model in the provider settings and make sure Tool use is ticked under capabilities. Check usually ticks it for you.
3

Chat in Work mode

In the desktop app, switch to Work mode and pick the model. Under OpenAI API Compatible, these models handle tool calls turn after turn: gpt-6.1-sol, glm-5.2, gemini-3.8-flash, kimi-k3.For Claude with tools, add the model to a provider set to Claude API Compatible (see Use Claude, Gemini, or Responses formats). Claude and DeepSeek can return a 400 when they call tools repeatedly under OpenAI API Compatible.
Since November 4, 2026, GPT-6 and GPT-5.6 models can reason and call tools in the same request, so gpt-6-sol and gpt-5.6-sol work with MCP too.

​
Model settings

Open a model in the provider settings to edit it:
  • Model Type: Chat, Image, Embedding, or Rerank
  • Capabilities: Vision, Reasoning, and Tool use
  • Context Window and Max Output Tokens: optional limits for the model

​
FAQ

​
The check fails or returns 404

Look at the preview URL. A 404 usually means /v1 is missing or doubled:
  • In OpenAI API Compatible mode, an API Host without /v1 plus an API Path of /chat/completions produces a URL with no /v1. Set the host to https://api.laozhang.ai/v1 and clear the path.
  • In Claude API Compatible mode, the host needs /v1; in Google Gemini API Compatible mode, it must not have it.
For more, see Invalid API key or 404.

​
I get a 401 error

The key is wrong or was copied incompletely. Copy it again from Tokens and confirm the token is enabled.

​
I get 403 or 503 “no available channels”

The token’s group doesn’t include the model, or the model ID is wrong. Use a default group token, or a Claude Code group token for a Claude API Compatible provider, and copy the ID from the model catalog. See Check model availability and access.

​
The connection still fails with the right address

Turn on Improve network compatibility in the provider settings and check again. In the web version, requests then go through Chatbox’s servers.

​
A model returns an error about temperature or Top P

o-series models and some Claude models don’t accept temperature or Top P. Set both back to unset in the chat settings and send again.

​
The thinking control is grayed out

The API mode doesn’t match the model. Add the model under the mode listed in Thinking controls and tick Reasoning in its capabilities.

​
The request fails with insufficient balance

See Why a request can fail with remaining balance. You can review each request in Logs.