NVIDIA NIM

Sythoria uses NVIDIA's OpenAI-compatible chat-completions endpoint for the NVIDIA NIM preset. Hosted requests are processed by NVIDIA under the account, model license, and service terms you accept.

Hosted API setup#

1. Sign in at build.nvidia.com.
2. Choose a model, accept any required license terms, and create an API key.
3. In Sythoria, open Settings, add a model, and select NVIDIA NIM.
4. Confirm the preset and paste the key.
5. Save and run the connection check.

Use the exact ID shown in NVIDIA's current catalog. Model availability and identifiers change independently of Sythoria. See the NVIDIA API catalog for current details.

Self-managed NIM#

For a NIM container you operate, create a Custom model and enter its complete chat-completions URL and model ID.

Loopback is appropriate for a service on the same machine. For another host:

• keep the endpoint on a private network;
• require authentication even if the container defaults to none;
• terminate TLS with a certificate trusted by the client;
• restrict source addresses and ports with a firewall;
• enable Sythoria local-network access only when the private address is intentional; and
• do not expose an unauthenticated inference port to the public internet.

Troubleshooting#

Do not publish the key, request body, private endpoint, or response in a support log.