Nvidia Provider

Adds NVIDIA NIM and self-hosted NIM microservices as a provider for the AI module. NVIDIA's hosted API is OpenAI compatible, so the provider gives you chat with streaming, tool calling and structured JSON output, plus embeddings, across NVIDIA's full hosted model catalogue.

Features

• Chat completions, including streaming, tool and function calling, and structured JSON output, through the AI module's OpenAI-compatible client.
• Embeddings, including NVIDIA's query and passage input types.
• Works with the hosted API at integrate.api.nvidia.com and with any self-hosted NIM microservice by changing the host.
• Lists every model your account can see and decides per model whether it is offered for tools, JSON, structured output or image input. Each guess can be corrected on the AI module's model settings form.
• Translates NVIDIA's error responses into readable AI module exceptions, so a retired or non-entitled model reports a clear message instead of failing deep inside the client.

Not supported: moderation, image and audio generation, which NVIDIA exposes as separate endpoints that are not OpenAI shaped.

Post-Installation

1. Create an API key at build.nvidia.com and store it as a Key entity.
2. Select the key at Configuration » AI » Providers » NVIDIA. Saving the form verifies the key by listing the models on the host.
3. Choose NVIDIA and a model for each operation at Configuration » AI » Settings.

The model list shows NVIDIA's whole public catalogue while an account is only entitled to part of it. A model outside your entitlement fails with "Not found for account"; pick another.

Additional Requirements

• AI module 1.5 or later
• Key module
• An NVIDIA API key, or a self-hosted NIM endpoint

Recommended modules/libraries

• AI API Explorer (part of the AI module) to try a model's chat, tool calling and embeddings before relying on it.
• AI Agents, if you want NVIDIA models driving agent loops. Tool calling quality and speed vary a lot between models, so test candidates first.

Community Documentation

See the README for configuration details, model capability handling and troubleshooting.