Home → Guides → Chatbox with a custom API: host, path, Claude and image generation

Chatbox with a custom API: host, path, Claude and image generation

Chatbox has three places where a custom API can go wrong: the API Host and API Path pair, the Claude mode that wants /v1 when Claude Code does not, and the Image Creator, which ignores custom providers entirely. We read the Chatbox 1.23.5 source, tested every combination against our gateway, and pulled the errors real Chatbox users hit.

Adding the provider

1. Open Settings → Model Provider and click Add Custom Provider. Give it a name and set API Mode to OpenAI API Compatible.
2. Enter https://apiclan.us as the API Host and leave API Path empty.
3. Paste your API key from the console.
4. Click Fetch to load the models of your key’s group, then Check to confirm the key works.

API Host	API Path	Chatbox calls
https://apiclan.us	(empty)	/v1/chat/completions ✓
https://apiclan.us/v1	/chat/completions	/v1/chat/completions ✓
https://apiclan.us	/chat/completions	/chat/completions ✗ 404

Chatbox adds /v1 to the host only when API Path is empty. Once you type a path, the host is used as it is — so https://apiclan.us with /chat/completions skips /v1 and returns 404 with an empty body. Either leave the path empty or put /v1 in the host.

Adding Claude

There are two ways, and the address rule differs between them:

API Mode	API Host	Chatbox calls
OpenAI API Compatible	https://apiclan.us	/v1/chat/completions ✓
Claude API Compatible	https://apiclan.us/v1	/v1/messages ✓
Claude API Compatible	https://apiclan.us	/messages ✗ 404

The simplest is a second custom provider in OpenAI API Compatible mode with a key from a Claude group — 29 of the 43 Chatbox requests we logged were claude-fable-5-1 sent exactly this way. If you prefer Claude API Compatible, the host must end in /v1, because Chatbox only appends /messages. That is the opposite of Claude Code, where ANTHROPIC_BASE_URL must not include /v1.

Generating images in Chatbox

The Image Creator only lists the built-in OpenAI provider (plus Chatbox’s own and Gemini). Custom providers are hard-coded to return no image model, so a custom provider fails with Provider doesnt support image generation.

1. Open Settings → Model Provider → OpenAI (the built-in one, not your custom provider).
2. Set the API Host to https://api.apiclan.us and paste a key created in an image group.
3. Add the model gpt-image-2.5 and set its Model Type to Image.
4. Open Image Creator and pick gpt-image-2.5.

Keep your chat models in the custom provider: a key belongs to one group, so the image key in the built-in OpenAI provider cannot run GPT chat models. We use api.apiclan.us here because a large image can take minutes and the main address cuts responses at 100 seconds. Our test request in Chatbox’s format returned a 1024×1024 image in 21 seconds.

What real Chatbox requests cost

All 43 Chatbox requests on our gateway in the 30 days to 12 October 2026, all streamed through /v1/chat/completions:

Model	Requests	Median first token	Average per message
claude-fable-5-1	29	5.2 s	$0.039
gpt-6-astra	8	1.4 s	$0.006
gpt-5.6-sol	6	7.6 s	$0.004

A Claude reply cost about seven times a GPT reply here, mostly because the answers were longer (622 output tokens on average). Chatbox resends the whole conversation every turn, but on Claude an average of 4,071 tokens per request were billed as cache reads instead of full-price input, which keeps long chats affordable. gpt-5.6-sol was cheapest but slowest to start, because it reasons before answering.

Errors real Chatbox users hit

• 404 with an empty body → an API Path without /v1 in the host, or Claude API Compatible without /v1. See the tables above.
• 400 This model is not supported on the Chat Completions endpoint → gpt-image-2.5 was chosen as a chat model. Image models only work in the Image Creator.
• 403 Image generation is not enabled for this group → the key in the OpenAI provider is not from an image group.
• 404 model_not_found (is not supported by any configured account in this group) → the model is not in your key’s group (one user typed gpt-5.2-pro, which we do not offer). Use Fetch instead of typing names.
• Provider doesnt support image generation → the Image Creator is pointed at a custom provider. Use the built-in OpenAI provider.

Other desktop clients: Cherry Studio with a custom API · Claude Code base URL.

Start using it

No subscription, no monthly minimum, no sales call. Top up with USDT and spend what you use — 1 USDT gives you 2 credits of API balance.

Read the 30-second quickstart

Language: English · 简体中文 · 繁體中文 · 日本語 · 한국어 · Deutsch · Français · Español · Türkçe · Português · Italiano · Bahasa Indonesia · Tiếng Việt · Bahasa Melayu · Polski · Čeština · فارسی · Русский

Prices on this page are filled in automatically from live billing data. Third-party terms are quoted from that party's own documentation.