Pi provider extension for Swiss AI Platform

Copyright (C) 2026, Daniel Roethlisberger.
https://github.com/droe/pi-provider-swiss-ai-platform

Experimental Pi extension enabling the use of models provided by Swisscom's Swiss AI Platform commercial inference service, offered through Swisscom's API platform Digital Marketplace.

Installation

pi install npm:pi-provider-swiss-ai-platform

Configuration

Interactive configuration as usual: /login -> «Swiss AI Platform» prompts for the subscription name, the Client ID and the Client Secret, verifies them against the gateway and stores them in ~/.pi/agent/auth.json. Access tokens are minted on demand and cached in memory until they expire.

Multiple configured subscriptions are not currently implemented. Switching subscription is another /login.

For headless use, the following environment variables can be set:

• SWISS_AI_PLATFORM_SUBSCRIPTION
• SWISS_AI_PLATFORM_CLIENT_ID
• SWISS_AI_PLATFORM_CLIENT_SECRET

Prerequisites

You need a Swiss AI Platform subscription, its technical name, and a matching Client ID and Client Secret.

For the subscription name, in your subscription on Digital Marketplace, select Documentation and check «Production Url». The last path component is the subscription name, e.g. all-models or apertus-1.5-70b. Note that for model-specific subscriptions, the subscription name is different from the model identifier.

For the credentials, in your subscription on Digital Marketplace, select «Credentials» and check the «OAuth 2.0 Credentials» section.

Models

As of September 2026, the following models offered as part of Swiss AI Platform work with Pi and show up in /model when using a subscription that includes them:

• google/gemma-4-31b-it
• mistralai/mistral-small-4-119b-2603
• qwen/qwen3.5-397b-a17b
• qwen/qwen3.6-35b-a3b
• swiss-ai/Apertus-v1.5-70B
• zai-org/glm-5.2-fp8

See MODEL_METADATA in index.ts for metadata on all models, including models hidden by this Pi extension, and the reason why they have been hidden. Feedback or patches to improve model compatibility very welcome.

Models added more recently, i.e. models not included in MODEL_METADATA yet, may or may not work, and likely need manual configuration in ~/.pi/agent/models.json.

Implementation Details

Models are served over the OpenAI-compatible /v1/chat/completions API; only some of them also support /v1/responses, so completions is the default and MODEL_METADATA.api opts a model into the Responses API. Authentication is an OAuth 2.0 client-credentials grant (HTTP Basic client authentication).

The model catalog is pulled from /v1/models on model refresh and cached by Pi in ~/.pi/agent/models-store.json. The endpoint returns ids only, so model capabilities come from the curated MODEL_METADATA table in index.ts. Unknown ids fall back to DEFAULT_METADATA; entries flagged hideModel: true are dropped from the catalog. Per-model overrides are still possible in ~/.pi/agent/models.json.

Known Issues

• Pi does not recognize [THINK]...[/THINK] thinking tags that mistralai/mistral-small-4-119b-2603 sometimes produces, unsure if this can be fixed with model metadata, would need updates to Pi proper, or is a failure mode of the model or the inference service.

Disclaimer

This Pi extension is inofficial, neither provided by nor supported by Swisscom.