Connect a Model

Point Orivael at any OpenAI-compatible endpoint — NVIDIA NIM, Ollama, vLLM, TGI, or a hosted API. Every prompt is run through the constitutional guard in and out. This is an internal console, so the API key is held server-side — you never paste it here.

Model Backend NOT CONFIGURED

🔒 Key is server-side. Set MODEL_API_KEY on the guard-api container (env). This page only sets the endpoint + model name — the key is attached on the server when a prompt runs, and is never sent to or stored in your browser.

NVIDIA NIM · deepseek-v4-pro NIM · llama-3.1-8b Ollama · llama3.2

Endpoint URL (OpenAI-compatible base) Model name

Test Run · Guarded Constitutional guard

Prompt