Pi Package for OVH AI Endpoints

Pi provider for OVHcloud AI Endpoints chat models.

Requires: Pi coding agent 0.81.0 or newer.

Installation

pi install pi-package-ovh-ai-chat

Setup

Get a token from endpoints.ai.cloud.ovh.net, then either:

# Option 1: environment variable
export OVH_AI_TOKEN="your-token"

# Option 2: stored credential (run inside pi)
/login ovhai

The extension registers even without a token, so you can install first and authenticate afterward.

Usage

# List available models (fetched dynamically from OVH)
pi --list-models | grep ovhai

# Chat with a model
pi --model ovhai/gpt-oss-120b

The model catalog is fetched from OVH on startup, on /model refresh, and on pi update --models. If the fetch fails (missing/expired token, OVH unreachable), the provider registers with an empty catalog and a warning; the next successful refresh populates it automatically.

Configuration

Variable	Required	Default	Description
OVH_AI_TOKEN	Yes	—	OVH AI Endpoints API token
OVH_AI_API	No	openai-responses	API type: openai-responses or openai-completions
OVH_AI_BASE_URL	No	https://oai.endpoints.kepler.ai.cloud.ovh.net/v1	Base URL override

Vision limitation: OVH's /v1/responses backend currently rejects image inputs (HTTP 422). For vision tasks, use a vision-capable model with OVH_AI_API=openai-completions:

export OVH_AI_API="openai-completions"
pi --model ovhai/Qwen3.5-9B

License

MIT