where to click Settings → Models: key, your own Base URL and your own model in the list
guide
How to paste the key
Pick an app and a model. If you’re in the cabinet, the key is already in the commands: copy, paste, done.
editors
Editors
Your key lives in model settings. If you only learn one — Cursor: gear at the bottom left → Models.
where to click Settings → Cascade / Model Providers → OpenAI Compatible
where to click settings.json → language_models.openai
where to click Settings → Models → Custom / OpenAI Compatible
where to click Settings → PearAI → API Key and Base URL
plugins
IDE plugins
Extensions for Cursor, VS Code, Neovim. Install from the marketplace. Provider is OpenAI Compatible — except Claude Code: two env vars, same as the terminal. Codex uses the same ~/.codex/config.toml as the CLI.
where to click ~/.claude/settings.json, or ANTHROPIC_BASE_URL + ANTHROPIC_AUTH_TOKEN
where to click Cline icon → gear → API Provider → OpenAI Compatible
where to click Roo settings → OpenAI Compatible
where to click settings → OpenAI Compatible
where to click file ~/.continue/config.yaml
where to click CodeGPT → Settings → API → Custom
where to click avante.setup → provider = openai
where to click ~/.config/opencode/opencode.json → provider + /connect
where to click ~/.codex/config.toml → model_provider + [model_providers.cheapai]
term
Terminal harnesses
Claude Code, OpenCode, Aider, Crush, Goose, Codex. Two variables, a json or toml file, then a command. Close the window and export is gone. Gemini CLI will not take this key — the card says so.
where to click ~/.claude/settings.json, or ANTHROPIC_BASE_URL + ANTHROPIC_AUTH_TOKEN
where to click OPENAI_API_BASE + OPENAI_API_KEY, then the aider command
where to click ~/.config/opencode/opencode.json → provider + /connect
where to click crush.json → providers.type = openai-compat
where to click goose configure → OpenAI, or OPENAI_HOST + OPENAI_API_KEY
where to click ~/.codex/config.toml → model_provider + [model_providers.cheapai]
where to click won’t start: Google only
desk
Desktop chat
A separate chat window, like your own ChatGPT. Watch the Base URL: some apps append /v1 and break.
where to click Providers → OpenAI-compatible
where to click Settings → API Host
where to click Settings → OpenAI → API endpoint
where to click Settings → AI Providers → Chat → Generic OpenAI
where to click Settings → Model Providers → OpenAI → Base URL
where to click Settings → Models → Add → OpenAI Compatible
where to click Settings → Providers → OpenAI Compatible
web
Chat in the browser
Your own front-end. The key lives in connection admin or yaml, not in every chat.
where to click Admin → Settings → Connections → OpenAI API
where to click librechat.yaml → endpoints.custom
where to click API → Chat Completion → Custom endpoint
where to click extension → Settings → OpenAI API
where to click Settings → Endpoint and API Key
auto
Automations
Put the key in Credentials / Model Provider, not in the scenario node. Otherwise every run is 401.
where to click Credentials → OpenAI → Base URL
where to click Settings → Model Provider → OpenAI-API-compatible
where to click Credentials → OpenAI API → Base Path
where to click OpenAI component → Base URL + API Key
code
Your own code
Same three fields. The guide has ready snippets — put the key in and run.
where to click base_url + api_key + model
nope
This key will not go in
ChatGPT.app, claude.ai, Copilot, Gemini CLI — their login only. Not a cabinet bug.
where to click won’t start
where to click won’t start
where to click won’t start