Connect a real model

The demo model only echoes. To get real answers, point Nuvora at a model endpoint you run or trust.

export NUVORA_PROVIDER_HOSTS='localhost,127.0.0.1,inference.internal.example'
export NUVORA_SECRET_VLLM='your-provider-key'
python3 -m nuvora.server

On k3s, pass the allow-list to the deploy script instead: NUVORA_PROVIDER_HOSTS=... ./scripts/deploy-remote.sh HOST USER.

In the console, Models → Add model offers presets for the Fabric AI gateway, Gryvia, vLLM and Ollama. Discover models from this endpoint then lists what the endpoint serves. See Zyvor platform integrations.

Adapters

Provider	Use it for	Notes
openai	vLLM, llama.cpp, Fabric, Gryvia, any OpenAI-compatible server	Supports tool calls; use it for agents
ollama	Native Ollama at http://127.0.0.1:11434, chat and /api/embed embeddings	For agents, use Ollama's /v1 endpoint with openai
aws	AWS-hosted models through Converse and ConverseStream	pip install '.[aws]'; standard AWS credential chain; no tools
demo	Offline evaluation	Synthetic, and always labeled

Rules the server enforces

• Allow-list. The host must exactly match an entry in NUVORA_PROVIDER_HOSTS. Redirects are refused.
• HTTPS. Remote hosts need HTTPS. Plain HTTP is accepted only for loopback.
• Secrets by reference. A model stores the name of a NUVORA_SECRET_* variable, never its value.

Routing

You can choose a provider explicitly in Playground, or use auto:

• auto picks the lowest configured input-plus-output price among the real chat providers that are enabled.
• It falls back to the demo only when no real provider exists.
• This is price-based routing, not a quality-aware router.