@howaboua/pi-gpt-switcher

Quick Pi commands for switching between GPT Codex models and reasoning levels.

Packages

Package details

extension

Install @howaboua/pi-gpt-switcher from npm and Pi will load the resources declared by the package manifest.

npm repohome report
$ pi install npm:@howaboua/pi-gpt-switcher
Package
@howaboua/pi-gpt-switcher
Version
0.1.5
Published
Oct 6, 2026
Downloads
933/mo · 133/wk
Author
howaboua
License
MIT
Types
extension
Size
23.9 KB
Dependencies
0 dependencies · 2 peers
Pi manifest JSON
{
  "extensions": [
    "./index.ts"
  ]
}

Security note

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

README

@howaboua/pi-gpt-switcher

Pi extension adding quick model commands:

Command Model
/sol [reasoning] openai-codex/gpt-6.1-sol
/luna [reasoning] openai-codex/gpt-6-luna
/astra [reasoning] openai-codex/gpt-6-astra

An explicit reasoning argument overrides the configured default. Valid arguments are off, minimal, low, medium, high, xhigh, and max; for example, /luna low.

Defaults

On first load, the extension creates pi-gpt-switcher.json in Pi's global agent directory:

{
  "sol": { "contextWindow": 272000, "reasoning": "high" },
  "luna": { "contextWindow": 472000, "reasoning": "xhigh" },
  "astra": { "contextWindow": 272000, "reasoning": "low" }
}

Edit the file to change a shortcut's session context window or default reasoning. Changes apply on the next shortcut invocation and do not alter the provider catalogue. Context windows must be safe integers of at least 128k. Values above the shortcut's default are accepted. Set a window supported by your provider; this setting changes Pi's budget, not the provider's capacity.

Existing config files need no migration: omitted shortcuts use their shipped defaults, and unknown entries are ignored.

Install

pi install npm:@howaboua/pi-gpt-switcher

Try it for one session:

pi -e npm:@howaboua/pi-gpt-switcher

How it works

The commands use Pi's model registry and normal provider authentication. If a model is unavailable or the OpenAI Codex credentials are missing, the command reports that instead of changing the current model.

Pi clamps the requested reasoning level automatically if the selected model supports fewer levels. GPT-6.1 Sol supports low, medium, high, xhigh, and max. An off or minimal request is clamped to low.

Local development

Load the checkout directly:

pi -e ./index.ts