ai_provider_nvidia 1.0.0-alpha1

First alpha of the NVIDIA provider for the AI module. It adds NVIDIA NIM and self-hosted NIM microservices as a provider, verified against the hosted API for chat, tool calling, structured JSON output, image input and embeddings.

What is included

• Chat with streaming, tool and function calling and structured JSON output, through the AI module's OpenAI-compatible client.
• Embeddings with NVIDIA's query and passage input types.
• Per-model capabilities: the provider decides from the model id whether a model is offered for tools, JSON, structured output or image input, and follows whatever you save on the AI module's model settings form.
• Readable errors: NVIDIA's RFC 7807 style failures are translated into normal AI module exceptions.
• Configurable host for self-hosted NIM deployments.

Requirements

• AI module 1.5 or later
• Key module

Known limitations

• The model list is NVIDIA's whole public catalogue; models your account is not entitled to fail with "Not found for account".
• Capability detection is name based. Correct a wrong guess per model on the AI settings model form.
• Moderation, image and audio generation are not supported.

Alpha means the API may still change and some rough edges remain. Please report problems in the issue queue.