Skip to main content
If you already pay for Claude Max, a ChatGPT plan that includes Codex, Google Antigravity, or xAI Grok, you can point CORE at your subscription instead of buying separate API credits. The CoreBrain Gateway includes a subscription proxy (CLIProxyAPI) that speaks OpenAI-compatible on the outside and translates to each subscription’s native protocol on the inside.
Subscription plans are licensed for interactive use. Providers can rate-limit or revoke tokens if they detect headless usage. Fine for personal self-host on a laptop or single VPS; risky if you’re serving other users. When in doubt, use real API keys.

​
The whole setup, four steps

​
1. Deploy the gateway

Pick one of Railway, Docker, or native. The gateway image bundles CLIProxyAPI, so nothing extra to install. On first boot it prints a public URL and a gwk_... security key.

​
2. Register the gateway in CORE

In the CORE webapp: Sidebar → Gateways → + New gateway. Paste the URL and security key.

​
3. Install the CLI and log in to your workspace

​
4. Log in to each subscription

One command per subscription. The CLI drives the OAuth flow end-to-end and lands tokens directly on the gateway.
Two flow types depending on the provider:
  • Claude / Codex / Antigravity — CLI prints an OAuth URL. Open it, approve, and the CLI’s local server catches the redirect and forwards it to the gateway. Nothing to type back.
  • xAI — CLI prints a URL and a short code. Open the URL, type the code on the page, done.
Add --gateway <name-or-id> if you have multiple gateways registered.

​
Wire the proxy into CORE

In the CORE webapp, open Workspace Settings → Models → API Keys, click + Add API key, choose OpenAI, and fill in:
  • Base URL — https://<your-gateway-host>/llmproxy/v1
  • API key — the same COREBRAIN_GATEWAY_SECURITY_KEY value from step 1
Then pick a model with the openai/ prefix. Examples: Get the exact list of available models by hitting the proxy’s model endpoint:

​
When a token expires

Just re-run the same command:
CLIProxyAPI auto-rotates access tokens as long as the refresh token stays valid; when it doesn’t, re-login. Tokens live on the gateway’s persistent volume, so a container restart never blows them away.

​
Structured outputs & tool calling

Under the hood, CORE routes structured-extraction calls (memory ingestion, label classification, etc.) via OpenAI-compatible tool calling. Claude’s grammar-constrained sampling makes this reliable even for empty content — the model will call the tool with empty arrays rather than respond with prose refusals like “there’s nothing to extract.”

​
Alternatives

The steps above use the built-in CLIProxyAPI inside the gateway. Other options exposing an OpenAI-compatible /v1/chat/completions surface work with the same BYOK wire-up:
  • Vercel AI Gateway — first-party pass-through for Claude Max via ANTHROPIC_BASE_URL. No self-hosting.
  • ai-cli-proxy-api — variant that supports Factory and Amp CLIs alongside ChatGPT Plus/Pro and Claude Pro/Max.