pi-auto-models

A pi extension that automatically switches between a primary and a fallback model based on quota. When your primary provider (Claude by default) is rate-limited, it transparently falls back (Codex by default) and switches back once quota recovers.

Features

• Auto-switch on quota — on each session start it uses the primary model if quota is available, otherwise the fallback.
• 429/529 recovery — detects rate-limit responses, caches the cooldown, and switches to the fallback mid-session.
• Passive quota tracking — reads rate-limit headers from provider responses (no extra requests for Claude); Codex quota is fetched live.
• /usage — shows each provider's quota windows (Codex windows are labelled by their actual duration, so it adapts if OpenAI drops the 5h window).
• /auto-model — interactive TUI to configure the primary/fallback model and thinking level.

Install

pi install npm:pi-auto-models

Or install from git:

pi install git:github.com/Fatpandac/pi-auto-models

Try it for one run without installing:

pi -e npm:pi-auto-models

Commands

Command	Description
/usage	Show Claude / Codex quota usage
/auto-model	Configure primary and fallback models (model + thinking level)

Configuration

/auto-model writes your choices to ~/.pi/agent/auto-model.json:

{
  "primary":  { "provider": "anthropic",   "model": "claude-opus-4-6", "thinking": "high" },
  "fallback": { "provider": "openai-codex", "model": "gpt-5.5",        "thinking": "high" }
}

Defaults (used when the file is absent):

Slot	Provider	Model	Thinking
Primary	anthropic	claude-opus-4-6	high
Fallback	openai-codex	gpt-5.5	high

Thinking levels: off, minimal, low, medium, high, xhigh.

How it works

State is persisted under ~/.pi/agent/:

• auto-model.json — your primary/fallback configuration
• claude-quota-cache.json — per-provider rate-limit cooldown expiry
• auto-model-rate-limits.json — last captured rate-limit snapshot per provider

Auth is read from pi's existing auth.json; the extension does not store credentials.