# fast

Priority service tier for OpenAI Codex in Pi.

## Install / load

Loaded through the root `pi-ext` package.
See [../../README.md](../../README.md).

## Commands / tools / settings

Commands:
none.

Tools:
none.

Hooks:

- `session_start`:
  refreshes settings from the session cwd and publishes status.
- `model_select`:
  refreshes settings from the session cwd and publishes status.
- `session_shutdown`:
  clears published status/footer segment.

Settings:

```json
{
  "fast": {
    "enable": true
  }
}
```

`fast.enable` is a boolean.
Trusted project settings override global settings; untrusted project settings are ignored.
A non-boolean value from the highest present source keeps fast mode off with a warning; lower sources are not consulted.
Default:
disabled.

## Behavior

Registers provider `openai-codex` with API `openai-codex-responses`.

When enabled, streams OpenAI Codex responses with native `serviceTier:
"priority"`.

When disabled, uses the normal simple OpenAI Codex response stream.

Scope:

- provider:
  `openai-codex`
- API:
  `openai-codex-responses`
- all OpenAI Codex models

On the priority path, reasoning options are clamped for the selected model
before
sending.
If the clamped reasoning level is `off`, no `reasoningEffort` is sent.

No model allowlist.

When active on `openai-codex`, publishes semantic `model-route:fast` state for `model-info` to render as the Nerd Font lightning-bolt icon `󱐋`.
Without `model-info`, it uses the icon as its status fallback.

## Debug

Opt in through [debug contract](../DEBUG.md).
Safe records: `session.start`, `session.shutdown`, and `settings.refresh` spans classified as `active`, `disabled`, `provider-mismatch`, `stale`, or publish error.