pi-openai-fast

Minimal Pi extension for OpenAI Fast mode.

It adds a /fast command and, when enabled, injects service_tier: "priority" into supported OpenAI/Codex provider requests. Codex uses priority as the wire value for the user-facing Fast tier.

Install

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

For local development from this checkout:

pi -e /home/dev/Development/pi-daniel/extensions/pi-openai-fast

Usage

/fast
/fast on
/fast off
/fast status

Fast mode is off by default and only affects the current Pi process. When it is on and the current model is supported, the extension publishes the footer status key pi-fast-mode with the text Fast.

Supported models

• openai/gpt-5.5
• openai/gpt-5.4
• openai/gpt-5.4-mini
• openai-codex/gpt-5.5
• openai-codex/gpt-5.4
• openai-codex/gpt-5.4-mini

The list is deliberately explicit so newly added OpenAI models do not start using the higher-usage Fast tier by surprise. Edit SUPPORTED_MODEL_KEYS in extensions/openai-fast.ts when OpenAI adds another Fast-capable model.

Behavior

The request payload is patched only when all of these are true:

• /fast is enabled.
• The active model is in the supported model list.
• The serialized provider payload is an object.
• The payload model matches the active Pi model id.
• The payload does not already contain service_tier.

There are no runtime dependencies, no config files, no provider registration, and no auth storage integration.

Development

npm install
npm run typecheck
npm test
npm pack --dry-run