Integration guides
Change the URL and the key. Nothing else.
Claude Code
Two environment variables: ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN. Claude Code speaks Anthropic’s format, so it runs the Claude models.
Cursor
Paste the key as the OpenAI API key and override the OpenAI base URL. Any model on the list works in chat and agent; Tab stays on Cursor’s own models.
OpenCode
One provider block in opencode.json with the openai-compatible package. The ids you list show up in /models.
Cline
Pick OpenAI Compatible, paste the base URL and key, type a model id and click Verify.
Roo Code
OpenAI Compatible, one model id at a time. Roo Code uses native tool calls only.
Kilo Code
A custom provider in VS Code or the CLI. Kilo fetches the model list from /v1/models for you.
OpenClaw
One provider block under models.providers in ~/.openclaw/openclaw.json, with api set to openai-completions.
SillyTavern
Chat Completion, Custom (OpenAI-compatible), the base URL and a key. Uncensored and DeepSeek models for roleplay.
Janitor AI
Provider Proxy, a new configuration with the chat completions URL, your key and a model id.
OpenAI, Anthropic and Gemini SDKs
Set OPENAI_BASE_URL or ANTHROPIC_BASE_URL and keep your code. Also the Gemini SDK and the Vercel AI SDK.
Images, video, speech and 3D
OpenAI’s image API for the image models, and one generations endpoint for the rest. Failed generations are not charged.
Nano Banana Pro
Price per image, and how to generate and edit with the OpenAI SDK or the generations endpoint.
Veo 3.1
Price per clip, next to Google’s, and how to submit and poll a generation.
What every guide shares
Each guide changes two things in your tool: the base URL and the key. The model ids, the prices and the balance are the same in every tool.
Base URL https://api.uplinkswitch.com/v1
Auth Authorization: Bearer sk-uplink-live-...
(x-api-key works too)
Models GET https://api.uplinkswitch.com/v1/models- One key from the console. The balance is prepaid from $10, with no subscription, and each top-up locks prices for 30 days.
- Three wire formats behind one URL. Every chat model takes the OpenAI format (
/chat/completions). The Claude models also take Anthropic's (/messages), and the Gemini models Google's. Each model on the Models page lists its formats. - A model costs the same in every format, so the Anthropic and OpenAI SDKs pay the same rate for the same Claude model.
- The key goes in a header. A key in the URL is refused.
- Every request shows on the console's Usage page with its tokens and its charge.
Which guide to follow
Follow the guide for the tool you already use. They differ in where the setting lives and in which format the tool speaks.
- Anthropic format only. Claude Code sends Anthropic requests, so it runs the Claude models. For GPT, Gemini, DeepSeek or Kimi in an agent, use one of the OpenAI-compatible tools below.
- OpenAI-compatible editors and agents. Cursor, Cline, Roo Code and Kilo Code take a base URL and a key in a settings panel. Every model on the list works, Claude included.
- Config files. OpenCode and OpenClaw read a provider block from a JSON file, so the setup can live in a repo or a dotfiles folder. Keep the key in an environment variable, not in the file.
- Roleplay. SillyTavern takes the base URL and key as a Custom (OpenAI-compatible) source; Janitor AI takes the full chat completions URL as a proxy. The uncensored builds run in both.
- Your own code. The SDK guide covers the OpenAI, Anthropic, Gemini and Vercel AI SDKs. The OpenAI and Anthropic SDKs read the base URL from an environment variable, so no code changes.
- Images, video, speech and 3D. Start with the media guide. The Nano Banana Pro and Veo 3.1 pages add per-unit prices.
Questions
Can I use one key in several tools at once? Yes. Every key on an account draws on the same prepaid balance. To tell tools apart, create a key per tool in the console and filter the Usage page by key.
How do I check a key before I open a tool? curl https://api.uplinkswitch.com/v1/models with the key in an Authorization: Bearer header lists the models it can call. Most tool guides have the exact command.
What happens when the balance runs out? Requests return 402 with the code insufficient_balance until you top up. The same message reaches whichever tool sent the request.