Client setup

One Nexus API key works in every client. Create it in the Telegram bot, then add a separate connection entry in the client so its own default stays unchanged. That entry is not a second Nexus account and does not create a second balance.

The base URL depends on the protocol the client speaks. Enter the base, not the full route. The client appends the operation path.

Workflow	Protocol	Base URL	Route the client builds
Chat apps and most OpenAI-compatible tools	OpenAI Chat	https://api.nexus-hub.tech/v1	POST /v1/chat/completions
Codex and response clients	OpenAI Responses	https://api.nexus-hub.tech/v1	POST /v1/responses
Claude Code and Claude Desktop	Anthropic Messages	https://api.nexus-hub.tech	POST /v1/messages
Gemini clients	Gemini-compatible	https://api.nexus-hub.tech/v1beta	POST /v1beta/models/{model}:generateContent

Setup order

1. Create a Nexus API key in the Telegram bot.
2. Add a separate connection entry in the client.
3. Choose the protocol and base URL from the table.
4. Put the key in the client's protected field or in an environment variable. Do not add a Bearer prefix unless the field asks for the whole header.
5. Copy a model id from GET /v1/models for this key. On Models and prices, check which route family that id belongs to.
6. Send a short message. A normal reply means the connection works.

Choose a protocol

OpenAI ChatOpenAI ResponsesAnthropic MessagesGemini-compatible

Use this for chat apps, including Cherry Studio. Base URL https://api.nexus-hub.tech/v1. The client calls POST /v1/chat/completions.

Use this for Codex. Base URL https://api.nexus-hub.tech/v1. The client calls POST /v1/responses. Do not switch Codex to Chat Completions.

Use this for Claude Code and Claude Desktop. Base URL https://api.nexus-hub.tech, with no /v1 suffix. The client calls POST /v1/messages and sends anthropic-version: 2023-06-01.

Use https://api.nexus-hub.tech/v1beta only when the client documents the Gemini format. The model id belongs in the request path.

Before troubleshooting

Recheck the route, the model id for this key, and the exact field in the client. Send support only the route, error code, approximate time, and x-request-id. Do not send the key, a full prompt, or request headers.

If it does not work

• 401 or invalid_key: the key is missing, has extra spaces, or was revoked.
• model_not_found or model_not_allowed: the id is not in this key's GET /v1/models list.
• unsupported_model: the id is visible, but not on this protocol. Pick another route or another id.
• endpoint_not_allowed: this key cannot use the selected route.
• 404 or unsupported_endpoint: compare the address the client actually calls with the table. Do not paste the full operation path into the base URL field.

Nexus also accepts a missing /v1, a repeated /v1, and a trailing slash. Keep the clean base from the table anyway, and trust the address preview when the client shows one.

The value table is on the compatibility matrix. Guides: Codex CLI, ChatGPT / Codex Desktop, Claude Code, Claude Desktop, VS Code, Zed, T3 Code, CC Switch, Cherry Studio, Direct API, custom integration, and another client.