@aliaksei-raketski/pi-fast-mode

A Pi extension that enables fast mode for supported models with one command and shortcut:

• /fast: toggles fast mode on/off.
• F3: toggles fast mode on/off.
• No arguments. /fast always toggles.

The current model's API and ID determine what gets injected. Any provider that speaks the model's API is supported, including proxies such as LiteLLM that prefix IDs with anthropic/ or openai/.

• Claude Opus 4.6 / 4.8 / 5 / 5.5 (anthropic-messages)
  • Adds speed: "fast"
  • Adds required header anthropic-beta: fast-mode-2026-02-01
• GPT-5.4 / GPT-5.5 / GPT-5.6 Luna, Sol, and Terra / GPT-6 Astra, Sol, and Luna
  • Adds service_tier: "priority"
  • openai-codex-responses requires ChatGPT/OAuth auth (API-key models are skipped)
  • openai-responses works with API-key auth

Install

pi install npm:@aliaksei-raketski/pi-fast-mode
# or project-local
pi install -l npm:@aliaksei-raketski/pi-fast-mode

Try locally from the repository root:

pi -e ./packages/fast-mode

This package ships the extension as TypeScript source. Pi loads extension entrypoints with its runtime TypeScript loader, so no JavaScript build output is required for this package.

Package contents

• extensions/fast-mode/index.ts Pi extension entrypoint.
• Fast-mode request hooks and session state helpers.

Behavior

• Start Pi with fast mode enabled:

pi --fast

• Footer status is always visible as one of:

  • fast on (accent color)
  • fast off (gray)

When fast mode is enabled but the current model is not supported, status is:

• no fast in the warning color (enabled, but inactive for the current model)

When you switch to a supported model, fast mode is applied automatically if it is enabled.

The fast mode toggle is stored in the current session, so it survives /reload, resume, and branch navigation.