Before you start
You’ll need:- A terminal on macOS, Linux, or Windows
- Node.js 18+ installed (required to run the Corti CLI via
npx) - A Corti Console account (free to create)
Install a coding agent
The Corti CLI supports the following agents. Install one (or more) using your preferred method.
- OpenCode
- OpenCode Desktop
- ForgeCode
- Crush
- Pi
Run the setup wizard
Run the interactive setup to authenticate and configure the Corti provider for your installed agents:The wizard opens by asking how you want to sign in:Press Enter to accept the default, Log in with a browser. The CLI shows a one-time code and opens the Corti Console. Approve the sign-in there, and the terminal continues on its own.If the browser can’t open automatically, for example over SSH, the CLI prints a URL for you to open instead.
Once you’re signed in, the wizard:
npx requires Node.js 18+. Download it from nodejs.org or install via a version manager like fnm or nvm.Paste an API key instead
Paste an API key instead
- In the Console, open Corti Models and generate your personal API key.
- In the CLI, choose Paste an API key and paste it.
CORTI_BEARER.Client credentials still work as well, mainly for headless and CI runs.- Verifies the connection and probes
/modelsso you can pick a default model. - Detects installed agents and pre-selects them (toggle with Space).
- Lets you choose scope: Global (
~/.config), Project (current directory), or both. - Shows a plan and asks you to confirm before writing anything.
Where the CLI writes your config
Where the CLI writes your config
The CLI writes a
corti provider entry into each selected agent’s config file, plus a managed .env block with your credentials:- ForgeCode —
~/.forge/.forge.toml - OpenCode —
~/.config/opencode/opencode.json(or.jsonc) - Crush —
~/.config/crush/crush.json - Pi —
~/.pi/agent/models.json
Load your credentials
If you let the CLI update your shell startup file, open a new terminal and skip ahead.Otherwise, load the
.env manually. Global scope writes it to ~/.env, project scope to .env where you ran the CLI.- macOS / Linux
- Windows
Next steps
AI Coding Tools
Prompt Claude Code, Cursor, Codex, or Lovable with a Corti build skill.
API Reference
Browse the full Corti API with interactive examples.
JavaScript SDK
SDK reference — the foundation the CLI and skills build on.
Agentic Framework
Multi-expert orchestration, MCP servers, and the agent runtime.
AI-generated code should always be reviewed and tested before use in
production. Human oversight ensures correctness, security, and compliance with
applicable regulations.