pi-codex-fast-mode

A pi extension that enables Codex fast mode by default and lets you switch between fast and standard modes.

It adds service_tier: "priority" to requests for models using pi's openai provider with the openai-responses API or the legacy openai-codex provider with the openai-codex-responses API. This avoids a model-ID allowlist that would need updates whenever OpenAI adds or renames a model. Models that do not support priority processing may ignore the setting or return an error.

Controls

/fast on
/fast off
/fast status

Fast mode defaults to on until you change it. The selected mode is saved globally in ~/.pi/agent/codex-fast-mode.json, so it applies to new sessions and survives reloads and resumes. Existing session-only preferences are migrated automatically. Supported models show ⚡ Codex fast or ○ Codex standard in pi's status bar.

Shared status key

The indicator is published under the CODEX_FOOTER_STATUS_KEY (codex-custom-footer) status slot. It appears alongside the separate quota status from codex-usage in pi's default footer and in compatible custom footers such as git-status.

Install

From npm:

pi install npm:@gowthamgts/pi-codex-fast-mode

From this repository:

pi install ./extensions/codex-fast-mode

Or try it without installing:

pi -e ./extensions/codex-fast-mode/index.ts

Use the openai provider (or legacy openai-codex) with ChatGPT sign-in to get Codex's credit-based fast mode. Direct API-key usage with openai applies API Priority processing and its separate token pricing instead.

Cost warning

Fast mode increases supported model speed by about 1.5x and consumes credits at a higher rate. As documented by OpenAI, GPT-5.5, GPT-5.6, and GPT-6 Luna/Astra/Sol use 2.5x Standard credits where available, while GPT-5.4 uses 2x.

See OpenAI's speed documentation.

Development

From this extension's directory:

pnpm test
pnpm run check:load