Claude Desktop App

The official Claude Desktop app does not expose API-key fields in its normal settings — but it supports a third-party inference mode that routes all model traffic (Chat, Cowork and Code) through a custom gateway. Enable it with the hidden Developer options.

Requirements: your gateway must implement the Anthropic Messages API (POST /v1/messages with streaming) — Nyxos API does. The gateway URL must be HTTPS.

Installation: download the official Claude Desktop app for macOS or Windows from claude.com/download.

---

Step-by-Step ConfigurationCopy Anchor Link

1. Enable Developer ModeCopy Anchor Link

1. Open Claude Desktop.
2. Go to Help → Troubleshooting → Enable Developer Mode.
3. Fully quit the app (Cmd/Ctrl+Q — closing the window is not enough) and relaunch it.
4. A new Developer menu appears in the menu bar.

2. Configure Third-Party InferenceCopy Anchor Link

1. Open Developer → Configure Third-Party Inference….
2. In Connection, set Inference provider to Gateway.
3. Fill in the gateway credentials:

4. In Models, add the exact Claude model IDs you want to use from /pricing (or leave discovery on if your plan exposes GET /v1/models).
5. Click Test connection to validate the URL and key.

3. RelaunchCopy Anchor Link

Claude Desktop reads this configuration once, at launch — fully quit and reopen the app after saving. New conversations now route through Nyxos API.

---

NotesCopy Anchor Link

• Account behavior: in third-party mode the app no longer signs into a Claude.ai account for inference — your sk-... key is the credential. Your key's group must include the models you add (use the automatic Auto-A group — see Groups & Multipliers).
• macOS quirk: if the app still signs in to Anthropic after reopening, set deploymentMode to 3p in the third-party configuration file, then quit and reopen again.
• Looking for MCP tool plugins instead? Those live in claude_desktop_config.json and are unrelated to model routing.