Pi fast mode extension

Install it with:

pi install npm:@valdo766hi/pi-fast

View this package on npm

This extension registers --fast and /fast, which opt the current session into OpenAI's priority service tier.

Commands

Start with fast mode enabled:

pi --fast

Or toggle it inside Pi:

/fast       # toggle
/fast on    # enable
/fast off   # disable

State is in-memory by default. Every new or resumed session starts off, so a session never inherits a paid tier from an earlier one. To explicitly opt a process or spawned Pi session in, start it with pi --fast or PI_FAST=1 pi; otherwise enable /fast on separately in that session. The footer shows a nerd-font bolt (nf-fa-bolt, U+F0E7) followed by FAST: ON or FAST: OFF.

The glyph is a private-use codepoint, so fast.ts writes it as the escape "\u{f0e7}" rather than a literal — editors and pipelines have silently replaced the literal with a space. fast.test.ts pins the codepoint.

Provider behavior

While fast mode is on, the before_provider_request hook adds "service_tier": "priority" to the request payload, but only when all of the following hold:

• the active model's provider is openai or openai-codex
• the active model's api is openai-responses or openai-codex-responses, the two apis that serialize a service_tier field
• the payload is a plain object that does not already carry service_tier

Everything else — other providers, OpenAI-compatible third parties such as groq or openrouter, Azure, and any explicitly set tier — is left unchanged. For Codex, costs are recalculated from the active model's base prices and given Pi's priority multiplier exactly once, whether the response reports default or priority. Responses are not charged twice when Pi already adjusted them.

Validated with Pi 0.85.1, 1.0.0, and 1.1.0 on Node.js 22.19+; Pi supplies the coding-agent and AI peers. See CHANGELOG.md for release notes.

Tests

node --test packages/fast/fast.test.ts

The same test runs in the repository's Node.js and devenv checks.