@benvargas/pi-openai-fast

/fast toggle for pi that enables OpenAI priority service tier on configured models.

This extension does not change the model, thinking level, tools, or prompts. It only adds service_tier=priority to provider requests when fast mode is active and the current model matches the configured supported-model list.

Requires pi 0.74.0 or newer.

Install

pi install npm:@benvargas/pi-openai-fast

Or try without installing:

pi -e npm:@benvargas/pi-openai-fast

Usage

• /fast toggles fast mode on or off.
• /fast on explicitly enables fast mode.
• /fast off explicitly disables fast mode.
• /fast status reports the current fast-mode state.
• --fast starts the session with fast mode enabled.
• By default, fast mode persists across new pi sessions via a JSON config file.
• Startup state comes from the selected config file, not from resumed session/thread history.

Example:

pi -e npm:@benvargas/pi-openai-fast --fast

Config

Config files follow the same project-over-global pattern as the other packages:

• Project: <repo>/.pi/extensions/pi-openai-fast.json
• Global: <agent dir>/extensions/pi-openai-fast.json

The global path uses pi's agent directory — typically ~/.pi/agent, but it follows PI_CODING_AGENT_DIR when that is set, matching where pi itself reads and writes config.

Releases before this one always used ~/.pi/agent, even with PI_CODING_AGENT_DIR set. If the agent directory differs from ~/.pi/agent and has no config yet, an existing ~/.pi/agent/extensions/pi-openai-fast.json is copied there once so previous settings carry over; a config already present in the agent directory is never overwritten.

If neither exists, the extension writes a default global config on first run.

Default config:

{
  "persistState": true,
  "active": false,
  "supportedModels": [
    "openai/gpt-5.4",
    "openai/gpt-5.4-mini",
    "openai/gpt-5.5",
    "openai/gpt-5.6-sol",
    "openai/gpt-5.6-terra",
    "openai/gpt-5.6-luna",
    "openai-codex/gpt-5.4",
    "openai-codex/gpt-5.5",
    "openai-codex/gpt-5.6-sol",
    "openai-codex/gpt-5.6-terra",
    "openai-codex/gpt-5.6-luna"
  ]
}

Settings:

• persistState: when true, /fast writes the current on/off state to config so it resumes in new pi sessions. Default: true.
• active: persisted fast-mode state used on startup when persistState is enabled.
• supportedModels: list of provider/model-id strings that should receive service_tier=priority.

Project config overrides global config. /fast on and /fast off write to the selected config file, so if a project config exists the remembered state is project-specific. If fast mode is enabled on a model that is not in supportedModels, the setting stays on but requests are left unchanged until you switch back to a configured model.

Notes

• When persistState is enabled, the last /fast setting also carries across brand-new pi sessions.
• Resumed sessions do not override the config-backed startup state.
• On configured models, fast mode maps to OpenAI service_tier=priority.
• openai/gpt-5.4-mini is in the defaults for the API-key provider only: OpenAI publishes Fast-mode pricing for it, but the ChatGPT (openai-codex) side offers no fast tier for gpt-5.4-mini. gpt-5.4-nano, gpt-5.4-pro, and gpt-5.5-pro have no published Fast-mode pricing and are left out.
• If a default set from an older release is found in your config, it is upgraded to the current defaults automatically; customized supportedModels lists are left untouched.

Uninstall

pi remove npm:@benvargas/pi-openai-fast

License

MIT