Integrations
Custom OpenAI Endpoint
How to get your Custom OpenAI Endpoint API key and connect it to Nova Admin.
Connect any OpenAI-compatible endpoint — vLLM, Ollama, LM Studio, llama.cpp server, TGI, Azure OpenAI, or any internal inference gateway that speaks the OpenAI Chat Completions API at /chat/completions.
Prerequisites
- A Custom OpenAI Endpoint account. Sign up at https://platform.openai.com/docs/api-reference/chat.
- An active Nova Admin workspace. Go to /agents/integrations to connect.
Steps
- Enter the base URL up to /v1 — e.g. https://vllm.internal:8000/v1
- Optionally provide an API key (some self-hosted endpoints accept any value)
- Set the default model name your endpoint serves (e.g. llama-3.3-70b-instruct)
- Set 'Allow Private IPs' to 'yes' only if your endpoint is on a trusted internal network (10.x, 172.16/12, 192.168/16). Loopback and cloud metadata endpoints stay blocked regardless.
- Use the Chat or List Models actions in flows; the model name can be overridden per-action
Credential fields
| Field | Required | Description |
|---|---|---|
| Base URL | Yes | url (e.g. https://your-endpoint.internal/v1) |
| API Key | No | Secret — stored encrypted, never exposed. (e.g. Leave empty if endpoint requires no auth) |
| Default Model | Yes | string (e.g. llama-3.3-70b-instruct) |
| Allow Private IPs | No | string (e.g. yes — to enable private/internal addresses (10.x, 172.16/12, 192.168/16)) |
Troubleshooting
- Invalid credentials: verify the key was copied in full with no trailing spaces.
- Key not working: check that your Custom OpenAI Endpoint account has API access enabled and billing is active if required.
- Need help?: refer to the Custom OpenAI Endpoint documentation.