Prerequisites
Clients
APIs and capabilities
Before you start: account, key and cost
Register, sign in, then create a key in key management. Choose its group deliberately: a model must belong to that group, and some groups also restrict the client. Check live pricing before use; this guide does not freeze prices or multipliers.
Choose a connection path
| Goal | Guide | Check first |
|---|---|---|
| Prepare a terminal | Node.js | node/npm on Windows, macOS or Linux |
| Work in a project with Codex | Codex | Responses protocol and group access |
| Use Claude Code | Claude Code | Anthropic root URL and model |
| Use Gemini CLI | Gemini CLI | Native Gemini protocol and authentication |
| Add a model in a GUI | TRAE SOLO | Custom provider protocol support |
| Configure an independent agent | OpenClaw, Hermes | An external-use group is required |
| Integrate your own program | API scripts | Protocol, response format and errors |
| Generate or edit images | Image generation | Image model and eligible group |
| Make structured judgments | TypeSafe | Currently usable entry: jev-latest |
Endpoint and group rules
The OpenAI-compatible base is https://monkeytheme.com/v1. The Anthropic base is https://monkeytheme.com; the client appends /v1/messages. Native Gemini routes use /v1beta/. Match the protocol and avoid duplicated version paths.
Codex Pro(仅限Codex) permits only ChatGPT Desktop and Codex CLI: no OpenClaw, Hermes, external scripts or image generation. Codex Pro(外接版) permits external clients and lists image models. Codex Lite(特价福利) permits external use but has no free Fast tier and currently lists no image models. Claude Max(仅限CC) permits only Claude Code and Claude Desktop. For external clients, choose Claude Max(外接版), Claude Plus(精品) or a suitable official direct group.
Perform a controlled verification
Check the client version, URL, exact model ID and key group first. A short request from your terminal can then verify access and may incur charges. Launching a client alone does not prove connectivity. No example here has been validated by live inference. Keep a redacted error code and request ID; do not share authorization headers, keys or complete private logs.
Platforms and capability limits
Use PowerShell on Windows, or WSL when a client requires Unix tools; use Terminal on macOS and your distribution's package manager on Linux. Follow each guide’s official installation and custom-provider setup steps. A model listing does not prove support for every tool, cache, image or video parameter. MonkeyAI has no Muse/Hunyuan groups; this documentation does not promise those capabilities.