@aarvay/pi-synthetic-provider

pi extension that registers Synthetic as a custom model provider.

Installation

Try it once without installing:

pi -e npm:@aarvay/pi-synthetic-provider

Install permanently into your user settings:

pi install npm:@aarvay/pi-synthetic-provider

Or install into a project (.pi/settings.json):

pi install -l npm:@aarvay/pi-synthetic-provider

Features

• Dynamically discovers available models from Synthetic's public /models endpoint at startup
• Falls back to hardcoded documented models if the fetch fails
• Supports text and image input models
• Tracks per-token pricing for cost estimation
• Handles reasoning/thinking levels
• Normalizes context overflow errors for automatic compaction and retry

Authentication

Synthetic requires an API key for inference. pi resolves the key in this priority order:

1. ~/.pi/agent/auth.json entry:

   {
     "synthetic": {
       "type": "api_key",
       "key": "syn-..."
     }
   }
   

2. SYNTHETIC_API_KEY environment variable

Models

The extension exposes all models returned by Synthetic's API. The documented aliases are:

• syn:large:text — Large text model (GLM-5.1)
• syn:small:text — Small text model (GLM-4.7-Flash)
• syn:large:vision — Large vision model (Kimi-K2.6)
• syn:small:vision — Small vision model (Qwen3.6-27B)

Use /model in pi to select a model.

Development

Test locally from the repo root without publishing:

pi -e ./packages/synthetic-provider

Troubleshooting

Models not loading: If the Synthetic API is unreachable at startup, the extension falls back to hardcoded models. Check your network connection with:

curl https://api.synthetic.new/openai/v1/models

API key errors: Verify your key is set via echo $SYNTHETIC_API_KEY or configured in ~/.pi/agent/auth.json.

License

MIT