Tool configuration
Point OpenAI Codex CLI at SeaLink and use Responses-compatible models as your terminal coding agent. What Codex can do — read files, run commands, make edits — depends on the permissions you grant it locally.
If your workload fits DeepSeek V4 Flash, you can try it through the same Responses-compatible endpoint. Set model to deepseek-v4-flash and keep base_url at https://api.sealink.io/v1. Before using it in automation, rely on the current model detail page and run a real codex exec check.
Requires Node.js ≥ 18. In your terminal: npm install -g @openai/codex Verify the installation with: codex --version Windows users: we recommend Windows Terminal. macOS and Linux users can use the built-in terminal.
Go to your SeaLink dashboard → Keys and either create a new API Key or copy an existing one (starts with sk-sealink-).
Codex reads the API key through the env_key specified in your provider config. Set it first: macOS / Linux: export OPENAI_API_KEY=$SEALINK_API_KEY Windows (PowerShell): [Environment]::SetEnvironmentVariable('OPENAI_API_KEY', $env:SEALINK_API_KEY, 'User') Close and reopen PowerShell for the change to take effect.
Edit ~/.codex/config.toml and add: model = "claude-opus-5-5" model_provider = "sealink" [model_providers.sealink] name = "SeaLink" base_url = "https://api.sealink.io/v1" env_key = "OPENAI_API_KEY" wire_api = "responses" supports_websockets = false
In your terminal: codex --help If you see the help text, the CLI is installed correctly. Then test with: codex exec "Reply exactly codex-ok" If Codex returns codex-ok without errors, it's working through SeaLink.
Navigate to your project folder. Run codex for interactive mode, or codex exec "your task" for one-shot execution. Codex can help you: - Read and analyze your codebase - Create and edit files - Run shell commands - Install dependencies
Env vars: Use PowerShell's [Environment]::SetEnvironmentVariable for permanent settings, or add them via System Properties → Advanced → Environment Variables. Terminal: Windows Terminal recommended. Node.js: Download the installer from nodejs.org.
Shell: Defaults to zsh, config file is ~/.zshrc. Older Macs may use bash with ~/.bash_profile. Node.js: brew install node is the easiest method. npm global path: Usually /usr/local/lib/node_modules/.
Shell: bash on most distributions, config file is ~/.bashrc. Node.js: We strongly recommend using nvm — apt's bundled version tends to be quite old. npm global path: ~/.nvm/versions/node/ if you're using nvm.
'codex: command not found'
npm's global bin directory isn't in your system PATH. Run npm config get prefix to find it, then add that path to your PATH. Common default locations: • Windows: %APPDATA%/npm/ • macOS / Linux: /usr/local/bin/ or ~/.npm-global/bin/
Codex still calls api.openai.com
Setting OPENAI_BASE_URL alone isn't enough. Open ~/.codex/config.toml and check that model_provider = "sealink" is set correctly, and that [model_providers.sealink].base_url points to SeaLink.
Slow or timed out
Codex reads your project context on first run, which takes longer than a simple chat request — that's expected. If timeouts persist, try testing in a small folder with a lightweight model first, then check your network connectivity and the status page.
Some models not compatible
Codex CLI currently requires Responses API semantics, and not all models support this. Use chat or code models that have been verified on the /v1/responses endpoint. For Claude models, use Claude Code via the Anthropic protocol instead.
