pi-yolo-auto

A pi provider extension for Yolo-Auto's flat-rate Qwen3.8-27B API

Adds an OpenAI-compatible provider with auto model-catalog sync and per-session subscription (Free/Builder/Pro) usage to pi.

Run

pi --model yolo-auto/qwen3.8-27b "hello"

Loads the provider from the pi package registry and runs one turn against the Yolo-Auto endpoint. The extension serves the embedded model catalog immediately and refreshes it from GET /models on session start, so the first real prompt already has a resolved model id.

Why pi-yolo-auto?

| | Capability | What it unlocks | |:--:|---------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------| | ⚡ | Flat-rate LLM access | A paid Qwen3.8-27B endpoint integrated like any native pi provider, with no per-token surprise. | | 🔄 | Self-updating model catalog | Stale-while-revalidate sync keeps models.json current without downtime or manual edits. | | 📊 | Subscription visibility | The status footer shows your tier (Free/Builder/Pro) and request counters straight from /usage. | | 🪟 | Plan-aware context | Qwen3.8-27B reports 128K context on Free/Builder and 256K on Pro, matching the site tiers; the catalog hot-swaps when the detected plan changes. |

How it fits

flowchart LR
    API1[/GET /models/] --> Sync[update-models.js]
    API2[/GET /usage/] --> Footer[(status footer)]
    Sync --> Models[models.json]
    Models --> Patch[patch.json]
    Patch --> Custom[custom-models.json]
    Custom --> Ext[index.ts provider]
    Ext --> Footer

The extension entry point (index.ts) is the composition root: it registers the provider, owns the stale-while-revalidate loop over the embedded catalog, and feeds the subscription footer. The pure pipeline (models.ts, usage.ts) stays dependency-free so it is unit-testable without pi-ai.

Available Models

| Model | Context | Vision | Reasoning | Input $/M | Cache Read $/M | Output $/M | |-------------|---------|--------|-----------|-----------|----------------|------------| | Qwen3.8 27B | 131K | ✅ | ✅ | $0.00 | $0.00 | $0.00 |

Install

Install it through pi's package registry:

pi install npm:pi-yolo-auto

Then load the extension and a yolo-auto provider appears in /model:

pi update --extensions   # or restart pi

Auth

Get your API key from your Yolo-Auto account, then log in from inside pi (recommended):

/login yolo-auto

You'll be prompted to paste the key; it is stored in ~/.pi/agent/auth.json (key in the access slot) and never printed back.

The /login path is the one and only auth path. No environment variable.

Usage

pi --list-models yolo-auto
pi --model yolo-auto/qwen3.8-27b "hello"

Credentials are read from ~/.pi/agent/auth.json (written by /login) and resolved at request time; they are never stored by this package or printed.

Documentation

• Configuration & data ownership: AGENTS.md
• Model catalog pipeline: models.ts
• Subscription parser: usage.ts
• Model cache & sync: scripts/update-models.js

License

MIT, © 2026 Ryan Brosas. Third-party licenses are listed in THIRD_PARTY_NOTICES.md when they exist.