pi-gpt

GPT fast mode for Pi, extracted from gpt-fast-mode.ts.

Run /fast to toggle priority processing for an explicit list of models. The extension adds service_tier: "priority" to matching requests and displays fast in the status bar. Provider support and account eligibility determine the actual processing tier and charges.

Install

Install from npm:

pi install npm:@likelz/pi-gpt

Or from GitHub:

pi install git:github.com/ekil1100/pi-gpt

Or from a local checkout:

bun install
pi install /absolute/path/to/pi-gpt

The package exposes TypeScript source through its Pi manifest, so local and Git installations work without a build step. Disable or move the original ~/.pi/agent/extensions/gpt-fast-mode.ts outside the extensions directory before loading this package to avoid registering /fast twice. Rename its JSON configuration to pi-gpt.json.

Configure

Edit ~/.pi/agent/extensions/pi-gpt.json (under PI_CODING_AGENT_DIR when overridden):

{
  "enabled": false,
  "models": ["gpt-6-astra"]
}

Entries support three forms, which can be mixed in the same list. Any matching rule enables fast mode when enabled is true:

• Model name without /: case-sensitive, exact match against the final segment of the model identifier. gpt-6-astra matches across providers, including magpie/codex/gpt-6-astra, and excludes gpt-6-astra-pro.
• Full identifier containing /: exact match against provider/model, including any additional model path segments.
• Regex wrapped in /.../ (without flags): JavaScript regular expression against the complete identifier. Existing regex rules keep their behavior. They use normal regex search semantics; add ^ and $ where you need boundaries.

Rule	Matches
gpt-6-astra	The exact final model name under any provider or path prefix
/[/]gpt-6-astra$/	openai/gpt-6-astra, openai-codex/gpt-6-astra, and magpie/codex/gpt-6-astra
/^openai[/]gpt-/	GPT model IDs under the openai provider
openai/gpt-5.4	Only this exact identifier
/.*/	Every model

[/] matches a literal slash, so the first rule requires the exact final model name and excludes gpt-6-astra-pro. Backslashes need JSON escaping: use "/gpt-5\\.4$/" to match a literal dot. Plain strings remain literal; * has regex meaning only inside /.../ rules. Invalid regex configuration produces a warning and starts with fast mode disabled.

The list starts empty; add rules for models whose endpoints support priority processing. Restart Pi or reload extensions after editing the file.

• /fast toggles and persists enabled.
• Unlisted models and disabled mode leave requests unchanged.
• Missing or invalid configuration starts with fast mode off and an empty list. Invalid configuration produces a warning.
• A failed save leaves the in-memory state unchanged.
• The status indicates the requested mode. It does not confirm the provider's actual processing tier.

Ultrafast

This extension requests priority, preserving the original behavior. OpenAI also documents service_tier: "ultrafast" for GPT-6 Astra in the Responses API. That is a separate tier with separate eligibility and pricing.

• API key: API billing and rate limits apply independently of ChatGPT subscriptions. The official guide says GPT-6 Astra Ultrafast is available to all API users at low limits; its published allowance table covers usage tiers 1–5.
• Codex subscription: Pro $500/month and eligible Enterprise/Edu plans have Ultrafast access. Other self-serve plans do not gain access by purchasing credits.
• A third-party proxy must support and forward the tier. This package does not verify account eligibility or enable Ultrafast.

Sources: Ultrafast API, Fast API, Codex speed and eligibility, Codex pricing.

Publishing

Publish to npm uses GitHub Actions OIDC. It needs no npm token secret. Node.js 24, npm 11.19.0, and Bun 1.4.2 are configured in the workflow. The prepublishOnly script runs checks, tests, and the build before publishing.

Configure a GitHub Actions trusted publisher in the npm package settings:

Field	Value
Organization or user	ekil1100
Repository	pi-gpt
Workflow filename	publish.yml
Environment name	Leave empty
Allowed actions	Allow direct publishing with npm publish

Run the workflow manually to check installation, validation, and tarball contents without uploading. This dry run does not verify npm's OIDC trust configuration; that is exercised by an actual release.

To release, update package.json to a new stable version, commit and push it, then push a matching tag such as v0.1.1. The workflow rejects tags that do not match the package version. Version 0.1.0 is already published and cannot be republished. OIDC publishing automatically includes provenance for this public repository and package.

Development

Use Bun 1.4.2 and Vite+ 1.0.0. Development was verified on Node.js 24.

vp install
vp check
vp test run
vp pack

vp pack builds the ESM library and declarations in dist/. vp run build runs the package's build script; vp dev starts Vite's application server and is not used for this extension. Pi supplies its host package at runtime; Effect is a runtime dependency.

The implementation uses pure functions for state validation and request transformation, Effect for file I/O and failure handling, and a small Pi adapter for commands and events. Tests use temporary directories and make no model requests.