# Choosing the AI

> Which AI CLI plays the prospect, which model, and how hard it thinks.

Every stage uses one AI CLI (the brain), one model and one effort level. Pass them as flags, or leave them out and pick from a menu.

## Brains

Three AI CLIs can play the prospect. The menus read their options live.

| --brain | Models listed from | Effort |
|---|---|---|

| `claude` | `claude --help` | low to max |
| `codex` | `codex --help` | low, medium, high |
| `opencode` | `opencode models` | Not asked; opencode has no effort setting |

```
leakdown your-site.com
```

Menus always offer **default** (the CLI's own setting) and, for models, **custom…** to type any id. Running bare `leakdown` opens a guided flow for all of it.

## In scripts and CI

With no terminal attached nothing is asked: the brain falls back to `claude` and the model and effort to the CLI's defaults. Pass all three so runs are reproducible. Each session records what it used in `meta.json`.

## Usage limits

A personal subscription can hit its usage limit on a long run. The queue stops, names the limit, and tells you to rerun later. Nothing is lost, and a ladder run resumes the same day.

Note

The CLI calls your AI CLI once per step, up to three attempts each. Automated use of a consumer subscription is between you and that provider's terms and rate limits.

---
Source: https://docs.leakdown.dev/choosing-the-ai