Codex Fast Mode

Pi extension that adds -fast model aliases for available models from the openai and openai-codex providers. Selecting an alias sends the original model ID with the OpenAI service_tier: "priority" request field.

Install

Install the umbrella Git package to load this extension alongside Plan Mode:

pi install git:github.com/pixelsnis/pi-extensions

To install just this extension from npm, after it has been published:

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

Restart Pi or run /reload after changing package settings. Remove it with pi remove npm:@pixelsnis/pi-codex-fast-mode, or remove the umbrella Git package with pi remove git:github.com/pixelsnis/pi-extensions.

What it changes

At session startup, the extension wraps the registered openai and openai-codex providers when they are available. For each catalogued model whose ID does not already end in -fast, it offers an alias with:

• ID <original-id>-fast
• Display name <original-name> (fast)

If the provider's model catalog already contains that alias ID, the extension skips it rather than creating a collision. The alias maps back to the original model ID for requests and adds service_tier: "priority" to the provider payload, while preserving an existing payload callback. Ordinary model IDs pass through unchanged. Providers other than openai and openai-codex are not modified.

The suffix is a model selector alias, not a promise that a request will be faster. Priority-tier availability, billing, rate limits, and eligibility depend on the provider and account. A provider may reject the field or the requested tier. Check current provider terms and usage settings before selecting an alias.

Verify without sending a request

Start Pi with the local extension, then inspect the model selector and look for -fast entries. Do not submit a prompt or select a fast alias just to test it:

pi --no-extensions -e ./extensions/codex-fast-mode

This check depends on the corresponding provider and model catalog being visible in your Pi installation. It does not require sending a model request.

Troubleshooting

• No -fast aliases: Check that the openai or openai-codex provider is registered and has models in its catalog. A model ID already ending in -fast is not aliased; a colliding alias ID is skipped.
• A priority request fails: The provider or account may not accept service_tier: "priority", or the requested tier may be unavailable. Retry with the original model ID and consult the provider's current documentation.
• Normal model requests behave differently: Only openai and openai-codex provider wrappers are installed, and only alias requests receive the priority payload transform. Report reproducible differences with the provider/model ID and Pi version.

Local development

From the repository root, load only this extension in an isolated local Pi session with pi --no-extensions -e ./extensions/codex-fast-mode. The source is index.ts; no build step is required because Pi loads TypeScript extensions directly.