# pi-multi-codex

A [pi](https://pi.dev) extension for using multiple OpenAI Codex accounts. It creates independent provider slots while leaving pi's built-in `openai-codex` provider unchanged.

It reuses pi's native OpenAI Codex provider, including its OAuth, model catalog, and streaming implementation.

## Requirements

- pi 0.82.0 or newer (native provider registration)

## Install

```bash
pi install npm:@khanhicetea/pi-multi-codex
```

For a one-off test:

```bash
pi -e ./index.ts
```

## Usage

Three slots are registered by default:

```text
openai-codex-1
openai-codex-2
openai-codex-3
```

Log into each account separately:

```text
/login openai-codex-1
/login openai-codex-2
/login openai-codex-3
```

Then use pi's built-in `/model` command to select an account. Slot models are labeled `[Codex 1] …`, `[Codex 2] …`, and so on.

Use pi's built-in `/logout` command to remove a slot's credentials.

### Fast mode

Run `/codex-fast` to toggle OpenAI's priority service tier for the current session. The footer shows `fast` while it is active, or `fast (inactive)` when fast mode is enabled but the selected model does not support it. You can also start pi with `--fast`.

Fast mode currently applies to these models on the built-in Codex provider and every numbered Codex slot:

- `gpt-5.4`
- `gpt-5.5`
- `gpt-5.6-sol`
- `gpt-5.6-terra`
- `gpt-5.6-luna`

To enable it by default, add the same setting used by `pi-codex-fast` to either the global `~/.pi/agent/settings.json` or project `.pi/settings.json`:

```json
{
  "pi-codex-fast": {
    "enabled": true
  }
}
```

For model-based opt-in, set a non-empty `fast_models` array instead. Fast mode then turns on automatically when selecting a listed model and off when selecting another model:

```json
{
  "pi-codex-fast": {
    "fast_models": [
      "gpt-5.4",
      "openai-codex-2/gpt-5.6-sol"
    ]
  }
}
```

A bare model ID matches that model across the built-in provider and every numbered slot. Use `provider/model` to target one provider slot. When `fast_models` is non-empty it takes precedence over `enabled`. A manual `/codex-fast` toggle remains in effect until the next model selection or session.

The project setting overrides the global setting. `/codex-fast` only changes in-memory state: it never writes settings or carries the toggle into another session. Edit the settings file to change the default across sessions.

### Usage checks

Run this command to fetch rate-limit usage for the account behind the currently selected Codex provider:

```text
/codex-usage
```

The result appears as a single-line compact widget, right-aligned below the editor beside the footer. It shows only the Codex slot name followed by each session or weekly window; account identifiers are omitted. Every window uses a color-highlighted remaining-quota bar with only the percentage left on the right. Model-specific limits are selected when the API exposes one for the active model.

The widget is displayed only when:

- `/codex-usage` is run manually, regardless of the remaining quota
- a background check after the agent's final settled turn finds any usage window below 20% remaining

Automatic checks are limited to once every five minutes per provider/model. Opening a session and switching models do not call the usage API and clear any previous widget. Automatic request failures stay silent; manual failures are reported. Usage credentials are resolved through pi's active provider, so each slot checks its own logged-in account.

## Configuration

Set `PI_CODEX_NUM_PROVIDER` to create between 1 and 100 slots (default: `3`):

```bash
PI_CODEX_NUM_PROVIDER=5 pi
```

Pi stores each slot's OAuth credentials under its provider ID. Browser and headless device-code login are supported.