Configure ZetaAPI in Claude Desktop
Connect Claude Desktop to ZetaAPI through its third-party inference (Gateway) mode and use Claude models in Cowork and Code without a Claude subscription, billed from your ZetaAPI balance.
Updated 2026-10-15
When to use this
Claude Desktop is Anthropic’s official desktop app. Its “Third-Party Inference” mode routes inference requests through a gateway compatible with the Anthropic Messages API. ZetaAPI’s /v1/messages endpoint is compatible with the Anthropic Messages API, so once configured you can use Claude models in both the Cowork and Code tabs, billed at ZetaAPI rates and without a Claude subscription.
Prerequisites
- Claude Desktop installed (if you can’t find the menu items below, update to the latest version first)
- An API key from the ZetaAPI console (starts with
sk-) whose group includes Claude models (the default group does)
Setup
1. Enable Developer Mode
From the app menu bar (macOS shown here; the menu bar is at the top of the screen), choose Help → Troubleshooting → Enable Developer Mode.
Note: this is the app menu bar, not the in-app Settings → Developer page. That page manages local MCP servers and has no third-party inference settings.
2. Open the third-party inference settings
A Developer menu now appears in the menu bar. Choose Developer → Configure Third-Party Inference….
3. Enter the gateway details
Fill in the fields as follows (field labels may vary slightly between app versions):
| Field | Value |
|---|---|
| Connection | Gateway |
| Base URL | https://api.staging.cosmos-router.com (do not append /v1) |
| API Key | Your ZetaAPI key |
| Auth scheme | Bearer (the default — no change needed) |
4. Save and restart
Click Apply Changes → Save & Restart. After the app relaunches, choose to continue in third-party mode on the sign-in screen (no Anthropic account sign-in needed). If Gateway appears next to your account name in the bottom-left corner, the switch succeeded.
5. Start using it
Switch to the Cowork or Code tab, pick a model (such as Haiku 4.5) from the model picker at the bottom right of the input box, and send a message.
FAQ
Q: Is prompt cache supported?
Yes. Multi-turn conversations in Cowork and Code hit the prompt cache, and cached tokens are billed at the cache price.
Q: How do I switch back to my official Claude account?
Choose to sign in with your Anthropic account on the sign-in screen. The third-party inference settings are stored only on this machine, so switching back does not affect your other devices.