CC Switch: connect Claude Code and Codex to OpenOxAI

Sign in and open API Keys

Sign in at openoxai.com/login, add credit to the plan you intend to use, then open openoxai.com/dashboard/keys.

Create the token and pick a Claude group

Click New Token, name it something you will recognise later (for example cc-switch-macbook), and choose the group that matches the workload.

Copy the key once

The full key is shown at creation time. Copy it straight into CC Switch in the next section — not into a shell profile, a README or a ticket. If you lose it, revoke the token and create a new one.

Open the Claude Code tab and add a provider

Make sure the Claude Code tab is selected in the icon row, then use the add control in the provider list. The dialog that opens is the one mapped below.

Fill the five fields

Name it OpenOxAI, paste the key, and use the root URL. Leave the advanced options alone unless your account needs a different auth header.

Save and check the card

The new card appears in the list with its name and note, but it is not active yet. If you keep several OpenOxAI keys, duplicate the profile and change only the key — the URL stays the same.

Click Enable on the OpenOxAI card

The button sits on the right side of the card. CC Switch writes the profile immediately; your previous profile stays stored and can be re-enabled later.

Confirm what was written

Open ~/.claude/settings.json. The profile's environment should be there — the OpenOxAI root and the token, with nothing pointed at Anthropic's own host.

"env": {
  "ANTHROPIC_BASE_URL": "https://api.openoxai.com",
  "ANTHROPIC_AUTH_TOKEN": "sk-…"
}

The real file usually contains more — a pinned model, permissions, hooks — depending on the profile. The two env entries decide where requests go.

Send a one-shot request

Start a new shell and run Claude Code with an explicit model. The exact IDs are the ones the model library lists.

claude --print --model claude-opus-5-5 \
  "Reply with exactly: CCSWITCH-OK"

--print sends the prompt, prints the answer and exits. Expect the single line CCSWITCH-OK. A hang is an auth or path problem, not a slow model.

Inside an interactive session, the same choice is /model claude-sonnet-5-5. The flag always wins over a model pinned in the profile.

Match the request in Usage

Open Usage and match the request by time and model. You should see claude-opus-5-5, a completed status and the token count for that turn. This catches the quiet failure: a reply billed to the wrong key.

Choose a GPT-service key

In OpenOxAI API Keys, create or select a key for the GPT service and confirm it can call gpt-6.1-sol. Use this exact lowercase model ID, not its display name. Do not use a Claude-service key for this profile.

Create a provider from the Codex tab

Select Codex in the top client row, click +, choose Codex Provider, then click Custom Configuration. Enter the provider name and GPT key. Keep API Request URL Full URL off and use this base URL:

Provider Name: OpenOxAI GPT
API Request URL: https://api.openoxai.com/v1
API Request URL Full URL: Off
Default Model: gpt-6.1-sol

Do not append /responses to the base URL. With Full URL off, Codex builds the Responses request path itself.

Set Responses (native), save and enable

Open Advanced Options and set Upstream Format to Responses (native). Confirm the model is gpt-6.1-sol, save with Add, then click Enable on the new profile. CC Switch fills the Codex auth.json credential field from the API key; never copy a real key into a screenshot, project file or shared document.

Restart Codex and send hi

Fully restart the local Codex client after enabling the profile, start a new session, choose gpt-6.1-sol and send hi. A model reply confirms the active client can reach OpenOxAI; a profile card by itself does not prove inference succeeded.

hi

Match the request in Usage

Sign in to OpenOxAI, open Usage, and match the request time and model. Confirm the GPT model ID, a successful status, token usage and cost. This is the final check that the request used the expected key and route.