Add Router One to Chatbox as a custom provider

Before you add the provider: which API mode for which model

Chatbox sends every model of a provider through that provider's API mode, so the mode decides which Router One IDs work. OpenAI API Compatible uses Chat Completions and covers every chat model, which makes it the only provider most people need. Add a second provider only when you want another wire format: OpenAI Responses API Compatible for GPT-family, DeepSeek V4 and Grok chat IDs, or Claude API Compatible for Claude-family and DeepSeek V4 IDs. Google Gemini API Compatible speaks Gemini's native API, which Router One does not serve; Gemini IDs work in OpenAI API Compatible mode. Create a key for Chatbox in Dashboard → API Keys with Create Key and give it a maxSpend cap; several Chatbox providers can share it.

Add Router One as a custom provider in Chatbox

Open Settings → Model Provider (设置 → 模型提供方 in the Chinese interface), click Add at the bottom of the provider list and choose Add Custom Provider. In the Add provider dialog, enter Router One as Name, choose OpenAI API Compatible as API Mode and click Add. On the provider page, enter https://api.router.one/v1 as API Host and leave API Path empty: Chatbox fills in /chat/completions, and the Preview line under the fields shows the full URL it will call, https://api.router.one/v1/chat/completions. In this mode Chatbox also adds /v1 to a bare host and moves a pasted /chat/completions into the path, so both mistakes correct themselves. Paste the dedicated key into API Key, then add models: Fetch loads GET /v1/models with your key, and New adds an exact ID by hand. The finished provider looks like this:

API 主机 / API Host

https://api.router.one/v1

# Chatbox：设置 → 模型提供方 → 添加 → 添加自定义提供商
# Chatbox: Settings → Model Provider → Add → Add Custom Provider
名称 / Name:          Router One
API 模式 / API Mode:  OpenAI API 兼容 / OpenAI API Compatible
API 主机 / API Host:  https://api.router.one/v1
API 路径 / API Path:  留空 / empty   → 预览 / Preview: …/v1/chat/completions
API 密钥 / API Key:   sk-your-router-one-key
模型 / Models:        获取 / Fetch，或新建 / New: anthropic/claude-sonnet-5

# Claude API 兼容 / Claude API Compatible（Claude 与 DeepSeek V4 的 ID）
API 主机 / API Host:  https://api.router.one/v1   → …/v1/messages（必须带 /v1 / /v1 is required）

Claude API Compatible mode: keep /v1 in the API host

In Claude API Compatible mode, Chatbox adds /v1 only when the host is exactly https://api.anthropic.com; for any other host it appends /messages to the host as entered. So enter https://api.router.one/v1, and the Preview line reads https://api.router.one/v1/messages. With the host root, Chatbox would call https://api.router.one/messages, which returns 404. That 404 message says Anthropic clients need the bare host, advice meant for Anthropic's SDKs, which add /v1 themselves; Chatbox's Claude mode does not, so keep /v1. Use this mode only for Claude-family and DeepSeek V4 IDs; another chat ID gets HTTP 400 with must be called via /v1/chat/completions. Fetch does not help here: in this mode Chatbox keeps only list entries marked with Anthropic's model type, which Router One's OpenAI-style model list does not carry, so add each ID with New.

Fetch, New and Check: what each button sends

In the two OpenAI modes, Fetch calls GET /v1/models with your key and lists the whole catalog, including image-generation models and IDs whose model page lists no capabilities. Add only the chat models you will use, or type an exact ID from /models with New, such as anthropic/claude-sonnet-5, openai/gpt-5.6-sol or deepseek-v4.1-flash. Check runs up to three real requests against a model: a plain one and, if that passes, one with an image and one with a tool call. All of them are billed and appear in Dashboard → Logs, and a failed image or tool test on a text-only or tool-less model is expected. Per Chatbox's guide, a model with no capabilities ticked is treated as text-only.

Model settings: capabilities, context window and output cap

Open a model's settings (Edit Model) to set what Chatbox may send it. Test Model runs the same checks and ticks Vision and Tool use when those requests succeed; you can also tick them yourself from the model page. Tool use matters beyond chat: per Chatbox's source, it attaches its tools for Agent Mode (MCP, skills, code execution), web browsing, knowledge bases and file reading only to models with Tool use ticked, and Agent Mode refuses a model without it (This model does not support Agent Mode). Set Context Window from the model page as a whole number, rounded down (1.05M becomes 1000000 for anthropic/claude-sonnet-5, and deepseek-v4.1-flash takes 1000000), leave Max Output Tokens empty unless you want a cap of your own, and tick Reasoning only for models that think.

Knowledge bases and images stay with other providers

A Chatbox knowledge base needs an embedding model to index files, and Router One serves no /v1/embeddings, so pick another provider's embedding model when you create one; the chat model that answers from it can still be a Router One model with Tool use ticked. Per Chatbox 1.23.5 source, custom OpenAI-compatible providers offer no image model, so the image-generation IDs that Fetch lists, such as gpt-image-2, cannot generate images in Chatbox; call them through /v1/images/generations from code or another client instead.

What Chatbox sends, and how to read its 4xx errors

Each message is one request, and with tools on, a reply can take several rounds, each billed; Check and Test Model add their own test requests. Filter Dashboard → Logs by the Chatbox key: each record shows the model, tokens, cost, status, total time and, for streamed requests that produced output, time to first token (TTFT). 401 means the key is missing or wrong; 404 usually means the API host does not fit the mode, most often the host root in Claude API Compatible mode; 400 with must be called via means the mode's endpoint does not serve that ID; 402 means the wallet balance or the key's maxSpend is used up.

Which model ID should Chatbox send?

Copy the exact model ID from /models, preserving case, hyphens, and version suffixes; do not substitute a display name. Open its detail page and match the supported API endpoints, context window, and capabilities such as tool calling to the provider and features selected in Chatbox. A catalog listing does not mean the client can use every feature of that model. Give each client or application a dedicated API key with a maxSpend cap.

Which API protocol is Chatbox using?

OpenAI-compatible describes an interface format; it does not make Chat Completions (/v1/chat/completions), Responses (/v1/responses), and Anthropic Messages (/v1/messages) interchangeable. Check the installed client version, provider configuration, and actual request path against the model detail page and API compatibility fact sheet. A successful plain-text chat does not establish support for hosted tools, conversation state, or file-editing features.

Verify the Chatbox call in your request trace

Send a simple text request from Chatbox, then match its trace in Dashboard → Logs by time, model, and request_id: tokens, cost, latency, and status. Next, test streaming, tool calls, and multi-turn history separately. For failures, retain the actual request path, full error message, and request_id. If there is no matching log, check client configuration and connectivity before attributing the error to the gateway or upstream.

FAQ