ElmonyDocs

Guides

Custom models

Add a provider that exposes an OpenAI-compatible API, then register the model identifiers you want Elmony to show in its model picker.

Before you begin

  • Confirm that the endpoint supports an OpenAI-compatible chat API.
  • Copy the API base URL and exact model IDs from the provider documentation.
  • Create an API key if the endpoint requires one. Keep credentials out of prompts and project source files.

Add a provider

  1. Open Settings and select Model Providers.
  2. Click Add custom provider.
  3. Complete the fields below, then click Add Provider.

Provider ID

A unique identifier without spaces. It cannot be edited after the provider is created, so choose a stable value.

Display Name

The provider label displayed in Settings and the model picker.

API Base URL

The provider's OpenAI-compatible base endpoint, commonly ending in /v1. Use the value from that provider's documentation rather than guessing it.

Environment Variable (optional)

The environment-variable name that contains the API key, for example MY_PROVIDER_API_KEY. You can instead save a key from the provider's Settings page when supported.

Register a model

  1. Open the new provider card under Model Providers.
  2. Save its API key if the endpoint requires authentication.
  3. In Models, click Add Custom Model.
  4. Enter the model fields and click Add Model.
  • Model ID: Must exactly match the identifier expected by the provider API.
  • Name: The friendly label shown in Elmony.
  • Description: Optional notes about the model's intended use.
  • Max Output Tokens and Context Window: Optional numeric limits. Use provider-documented values if you set them.

Select and test it

Return to Builder Chat, open the model picker, expand the custom provider, and select the model. Start with a small Ask or Simple request. A 401 or 403 response usually points to credentials or provider permissions; a 404 often means the base URL or Model ID does not match the provider API.

Compatibility note

"OpenAI-compatible" implementations are not always identical. Streaming, tool calls, structured output, image input, or token limits may differ. If normal chat works but a specialized mode fails, check that the endpoint supports the capability that mode uses.