https://app.trynito.ai/v1 as the API base URL, your Nito API key as the credential, and a model ID such as moonshotai/kimi-k3. Hermes saves the provider, URL, and model choice in its configuration for later sessions. Use the setup wizard to enter the key instead of writing it into a shared config file.
For Confidential inference, choose a catalog ID ending in :confidential. Hermes uses Nito’s Chat Completions endpoint for this custom provider.
Check the route
Starthermes chat and request a short reply. Then ask for a small task that uses a Hermes tool. Text success confirms the endpoint and key; a tool task checks whether the chosen model works with the agent loop. Inside a session, /model switches among providers already configured. Use hermes model in the terminal to add or change an endpoint.
Auxiliary tasks and troubleshooting
Hermes can use separate models for auxiliary tasks such as vision or summarization. Those tasks follow the Hermes auxiliary configuration; selecting Nito for chat does not automatically establish the route for every separately configured task. Choose the main provider for an auxiliary task when you want that task to use the same Nito route. If a request goes to the wrong host, inspect Hermes’ saved custom endpoint.OPENAI_BASE_URL applies to Hermes’ direct OpenAI provider in current Hermes versions; hermes model sets the custom endpoint used here. For a 404, check that the URL ends in /v1 and the model ID matches the catalog.
Hermes documents the interactive flow and its saved configuration in Custom and Self-Hosted LLM Providers.