Skip to main content
This page explains how to run models from Modular Cloud on Pi.

Before you begin

Make sure you have:

Configure a Pi provider

To register Modular Cloud as a provider in Pi:
  1. Open (or create) the ~/.pi/agent/models.json configuration file.
  2. Choose the model that you want from our list of shared endpoints or use a dedicated deployment if applicable.
  3. Add Modular Cloud as a provider:
    models.json
    Modular Cloud supports the OpenAI chat completions API, so set api to openai-completions. Modular Cloud doesn’t support the OpenAI responses API.
Once you save your models.json file, run /model in Pi and select the Modular Cloud model to use.
The example above demonstrates a basic provider and model configuration. To see the complete list of configuration options, see the Pi documentation on Custom Models.

Retry rate-limited requests

Shared endpoints can return a 429 error when you hit a rate limit or when the endpoint is at capacity. To have Pi pause and retry failed requests, add this to ~/.pi/agent/settings.json:
settings.json
  • maxRetries sets how many times Pi retries a failed request from the model provider before it stops trying.
  • maxRetryDelayMs sets the maximum server-requested delay in milliseconds. The default is 60000 (1 minute).
For more details, see the Pi settings reference.