pi-model-fast

Fast mode for Pi: per-model OpenAI/Codex priority controls and an auto-refreshing model catalog

Packages

Package details

extension

Install pi-model-fast from npm and Pi will load the resources declared by the package manifest.

npm repohome report
$ pi install npm:pi-model-fast
Package
pi-model-fast
Version
0.1.2
Published
Oct 3, 2026
Downloads
178/mo · 17/wk
Author
jiuai233
License
MIT
Types
extension
Size
31.9 KB
Dependencies
0 dependencies · 1 peer
Pi manifest JSON
{
  "image": "https://raw.githubusercontent.com/jiuai233/pi-model-fast/main/docs/assets/fast-status.png",
  "extensions": [
    "./src/index.ts"
  ]
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

pi-model-fast

Fast mode for Pi, with per-model OpenAI/Codex priority controls, colored status indicators, and an automatically refreshed model capability catalog.

简体中文

Fast enabled in Pi

Quick start

Requires Node.js 24+ and Pi 1.0.0+.

pi install git:github.com/jiuai233/pi-model-fast

Inside Pi:

/reload
/fast on

Fast is off by default. Settings are saved separately for each provider and model. Switching models does not enable Fast on an unconfigured model. Disable other Fast extensions to avoid competing request changes.

Fast may increase provider usage or cost. Account access, endpoint support, and provider rules still apply.

Features

  • Supports OpenAI API and Codex models declared Fast-capable in the official catalog.
  • Preserves the selected model, reasoning effort, and tools.
  • Sends OpenAI/Codex service_tier: "priority" when enabled.
  • Shows a colored lightning indicator with Fast ON; hides the indicator when off.
  • Refreshes model capabilities without using model credits or reading OpenAI credentials.

Commands

Command Action
/fast Toggle Fast for the current model
/fast on, /fast off Enable or disable Fast
/fast status Show the model's setting, capability, and last request failure
/fast list List Fast-capable models
/fast refresh Refresh the official capability catalog immediately

Status indicator

  • Green lightning + Fast ON: enabled and ready to apply Fast parameters.
  • Red lightning + Fast ON: Fast parameters cannot be applied, or the latest Fast request failed. A successful retry restores green.
  • Off: the indicator is hidden.

Manual enable/disable clears the current model's failure record. Cancelling a request is not counted as a failure. Turning Fast off stops this extension from injecting parameters; it leaves parameters set by other sources intact.

Catalog support does not guarantee account access or faster responses. Codex can echo service_tier: "default" even when the request used priority; that response field alone cannot verify acceleration.

Automatic model catalog

The capability source is the official OpenAI Codex model catalog, using service_tiers and additional_speed_tiers.

Models declaring priority or fast enter the Codex catalog. Models also declaring supported_in_api enter the OpenAI API catalog. Catalog additions do not enable Fast automatically.

  • A capability snapshot is bundled with the extension.
  • Startup and task submission check the cache; snapshots older than six hours refresh in the background.
  • Failed refreshes retain the last catalog and show a notification. Automatic retries are spaced six hours apart.
  • /fast refresh refreshes immediately.
  • GitHub Actions checks upstream every six hours and commits catalog changes.

Catalog refreshes access public GitHub data, make no model requests, and consume no model credits. Snapshots contain model names, identifiers, and providers; no upstream prompts or account data are retained.

Local settings

Files are stored under the Pi configuration directory's extensions/ folder, normally ~/.pi/agent/extensions/. PI_CODING_AGENT_DIR is respected.

  • pi-model-fast.json: per-model settings.
  • pi-model-fast-catalog.json: capability cache.

Example:

{
  "version": 1,
  "enabledModels": {
    "openai-codex/gpt-6.1-sol": true
  }
}

Development

npm ci --ignore-scripts
npm run check
npm run update:catalog

Tests use temporary directories and simulated extension events; they make no real model requests.

Provider adapters are in src/providers.ts; capability handling is in src/catalog.ts. New providers require an adapter, an official capability source, and corresponding tests.

Update or uninstall

pi update git:github.com/jiuai233/pi-model-fast
pi remove git:github.com/jiuai233/pi-model-fast

Run /reload afterward. Uninstalling leaves settings and cached capabilities intact.

References

License

MIT.