Why doesn't ~/.claude/settings.json change the Claude desktop app?

The desktop app reads its model route from its own third-party inference configuration, not from ANTHROPIC_BASE_URL or ~/.claude/settings.json. Those settings only affect Claude Code in the terminal and in editors. In the desktop app, use Developer → Configure Third-Party Inference.

Do I need a Claude Pro or Max subscription?

No. Once a gateway is configured, the sign-in screen offers to start the app with that configuration instead of signing in to Anthropic. Requests are paid from your LLMsRelay balance.

Which gateway base URL do I enter?

Enter https://api.llmsrelay.com exactly: no /v1 at the end and no trailing slash. The app adds the API path itself.

Should the gateway auth scheme be bearer or x-api-key?

LLMsRelay accepts both. Leave the default bearer value unless you have a reason to change it.

What stops working in third-party mode?

Remote Control (continuing a session from your phone or claude.ai/code) and Anthropic-hosted cloud environments are unavailable while the gateway configuration is active. Chat, Cowork, and Code work through LLMsRelay, and conversations are stored on your computer.

Claude Desktop App Setup with a Custom Base URL and API Key

Back to dashboard

LLMsRelay is an independently operated API gateway. It is not affiliated with or endorsed by Anthropic, PBC. Product and model names are used only to describe compatibility.

Before you start

The Claude desktop app for macOS or Windows, from claude.com/download. Update it if it is old: the third-party inference form is in current versions.
An LLMsRelay account with a balance. Add a one-time usage pack on Plans if the balance is empty.
About ten minutes. You do not need a Claude Pro or Max subscription.

Create an LLMsRelay key

Open API Keys in the dashboard.
Create a key in the Basic or Pro group. It starts with sk-cs4-.
Copy the key and keep the page open. The full key is shown only once.

Turn on Developer Mode

The gateway form lives in a hidden Developer menu. Developer Mode makes that menu appear. It does not change your files or chats.

macOS

Open the Claude app and click anywhere in its window.
In the menu bar at the very top of the screen, open:

HelpTroubleshootingEnable Developer Mode…

Windows

Open the Claude app and click the menu button with three lines (☰) in the top-left corner of the window.
Open:

HelpTroubleshootingEnable Developer Mode…

A dialog asks Enable Developer Mode? and mentions developer tools and debugging features. Click Enable. The app restarts by itself. After the restart a new Developer menu appears next to Help (on Windows, inside the ☰ menu).

If the app runs in another language, the menu names are translated but stay in the same places.

Open the third-party inference form

DeveloperConfigure Third-Party Inference…

A separate settings window opens. This is where the desktop app stores its model route: the address of the service it sends requests to and the key for that service.

Enter the LLMsRelay gateway

Fill in the fields exactly as shown. Leave every field that is not listed here empty.

Inference provider: Gateway
Gateway base URL: https://api.llmsrelay.com — no /v1 at the end and no trailing slash
Credential kind: Static API key (if the field is shown)
Gateway API key: Your key from step 1, starting with sk-cs4-
Gateway auth scheme: Leave bearer. x-api-key also works with LLMsRelay.

Gateway base URLtext

https://api.llmsrelay.com

Click Apply Changes, then Save & Restart. The app restarts by itself.

Start the app with LLMsRelay

After the restart the sign-in screen offers to start the app with the configuration you entered. Choose that option, not the Anthropic sign-in. You do not need to enter an email address or a password.

The app now runs as Claude (third-party). It is a separate mode: Chat, Cowork, and Code all work through LLMsRelay, conversations are stored on this computer, and chats and projects from a claude.ai account are not shown there.

Check that requests go through LLMsRelay

Open the Code tab and start a session in any folder.
Send a short test message:

Test messagetext

Reply with exactly: pong

Claude answers pong. In third-party mode the app has no other route to a model, so an answer already means the request went through LLMsRelay.
Open the LLMsRelay dashboard: the request appears in your usage within a minute.

For a technical check, type /status in the message box: the Anthropic base URL line shows https://api.llmsrelay.com.

The cost the desktop app shows is an estimate at Anthropic list prices. The actual charge is the one in the LLMsRelay dashboard.

Choosing a model

Pick the model in the model selector of the Code tab, as usual. If a model answers with an error such as model not found, choose another one, for example Claude Opus 5.5.

The models available to your key are listed on the Models page. You can also ask the gateway directly:

Models available to your keybash

curl https://api.llmsrelay.com/v1/models \
  -H "x-api-key: YOUR_SK_CS4_KEY"

What changes in third-party mode

Chat, Cowork, and Code work through LLMsRelay and are billed from your LLMsRelay balance.
Conversations are stored on this computer, not in a Claude account.
Cowork needs macOS 14 or later, or on Windows the app installed from the .msix package. Chat and Code work without it.
Remote Control is unavailable: you cannot continue these sessions from your phone or claude.ai/code.
Anthropic-hosted cloud environments are not offered in the environment selector.
The configuration is stored on this computer only. Repeat the steps on every computer where you want the same setup.

Going back to a Claude account

On the sign-in screen, choose the Anthropic sign-in option instead of the third-party one. The LLMsRelay configuration stays saved, so you can switch back the same way. To change the key or the address, open Developer → Configure Third-Party Inference… again at any time.

Developer Mode can stay on: it only adds the Developer menu and does not change how the app works with a Claude account.

Troubleshooting

There is no Developer menu

Developer Mode is off or the app has not restarted yet. Repeat step 2 and wait for the restart. On Windows the menu is inside the ☰ button in the top-left corner. If Help → Troubleshooting has no Enable Developer Mode item, update the app.

The sign-in screen has no third-party option

The app did not read the configuration. This happens when it was saved while the app kept running. Quit the app completely (Cmd+Q on macOS, Exit in the ☰ menu on Windows) and open it again. Also check that the inference provider is Gateway and that both the address and the key are filled in.

“Gateway was unreachable” or “Can’t reach your inference provider”

The app could not reach the address. Check that the gateway base URL is exactly https://api.llmsrelay.com. Open api.llmsrelay.com/healthz in a browser: if it does not open, a VPN, proxy, or firewall is blocking the connection.

“The configured API key was rejected by the inference provider”

Open the form again and paste the key once more, without spaces. Check that the key belongs to the Basic or Pro group and has not been revoked in API Keys.

Requests stopped because of the balance

Add a usage pack on Plans. The key keeps working after the top-up; nothing needs to change in the app.

The app still uses my Claude account

The app was started with the Anthropic sign-in instead of the third-party configuration, or the configuration was not saved. Open the form, check the values, relaunch, log out of the claude.ai account if the app opened it, and choose the inference provider option on the sign-in screen. Requests that do not appear in the LLMsRelay dashboard did not go through LLMsRelay.

The terminal works, the desktop app does not (or the other way round)

They are configured separately. The terminal and editors read ~/.claude/settings.json; the desktop app reads its third-party inference form. The same LLMsRelay key can be used in both places.

Ready to start?

Create a key and configure a compatible API route in under 2 minutes.

View Plans