hjanuschka/pi-multi-pass

What it solves

pi-multi-pass solves the problem of hitting API rate limits when using the pi coding agent. It allows users to manage multiple OAuth accounts for the same AI provider, automatically rotating between them when one account is exhausted to ensure uninterrupted workflow.

How it works

The extension manages a collection of subscriptions and organizes them into several routing mechanisms:

• Rotation Pools: Groups of accounts for a single provider. When a rate limit is hit, the system automatically switches to the next account in the pool and retries the prompt. Selection strategies include round-robin, quota-first (preferring accounts with more headroom), scheduled (time-based windows), and custom (via JS scripts).
• Fallback Chains: Ordered sequences of pools or models that the agent tries in succession if the primary pool is exhausted.
• Model Presets: Named shortcuts (e.g., coding-premium) that map to a preferred list of providers and models across different services.
• Project Affinity: Users can create a .pi/multi-pass.json file in specific project directories to restrict which accounts or pools are used for that project (e.g., separating work and personal accounts).

Who it’s for

Developers using the pi coding agent who have multiple subscriptions to AI providers (like OpenAI, Anthropic, or Google) and want to avoid downtime caused by rate limits.

Highlights

• Automatic Failover: Seamlessly switches accounts and retries prompts upon detecting rate-limit errors.
• Quota Monitoring: Built-in support to check remaining usage limits for Codex and Google providers.
• TUI Management: All configurations are handled via in-agent commands (/subs, /pool, /mp-preset) rather than manual config file editing.
• Flexible Routing: Supports complex logic for account selection, including time-of-day scheduling and custom JavaScript selector scripts.
• Project-level Isolation: Ability to whitelist specific subscriptions per project folder.