Add a custom provider

Use Models.dev or Local Servers in Settings to add a provider that is not one of Kora’s built-in entries.

Add a provider

1. Open Settings.

2. Pick a method and fill the form:

   Catalog

   Open Models.dev and start typing the provider name. Select a result, enter its API key, and save. Kora fills the base URL, models, and capabilities from the catalog.

   Manual

   Open Models.dev, choose the manual-add action, and enter a display name, OpenAI-compatible base URL, API key, and model definitions.

   Local

   Open Local Servers, choose a preset or manual configuration, and enter the server base URL. Click Discover from server to load GET /v1/models, then save.

3. Close Settings and select one of the new models in the chat composer.

Built-in providers

A few providers ship preconfigured so you can connect with just an API key.

Provider	Endpoint	Notes
Kimi (kimi-code)	https://api.kimi.com/coding/v1	OpenAI-compatible. Lists Kimi Code models by default. The minimum temperature is normalized to 1.0 for all Kimi-family models across any provider.

Kimi is also available as a manual entry under models.dev if you want to override the base URL or pin different models.

Dynamic model discovery

Some providers expose a live catalog endpoint. Kora fetches the catalog on demand and persists the result so the dropdown stays usable offline.

• ByteDance ModelArk — fetches GET /v1/models from your endpoint, shows each model with a per-model toggle, and exposes Test Connection in the provider detail page so you can confirm the route before sending a real message.

Per-model capabilities

After adding a provider, edit each model’s capabilities at any time from the provider detail view:

Capability	Effect
Tools	Enables function/tool calling
Streaming	Enables token-by-token streaming
Vision	Enables image input
Reasoning	Enables reasoning/thinking output

Per-model temperature

Each model can have its own temperature override. See Per-model temperature for the range and precedence rules. Kimi-family models are forced to a minimum of 1.0 regardless of the per-model override.

Troubleshooting

• Provider doesn’t appear after save → reload the model dropdown; the list is fetched asynchronously after save.
• Discover from server errors → the local server must be reachable from your machine and expose GET /v1/models. Try curl http://localhost:11434/v1/models for Ollama.
• API key rejected → some providers expect a Bearer prefix, some don’t. Paste only the token; Kora formats the header.
• Provider is saved but has no models → reopen its detail page and add a model or refresh the models.dev entry. For ModelArk, the catalog only refreshes when Test Connection succeeds.
• Kimi answers ignore low temperatures → this is expected — the minimum temperature is forced to 1.0 for every Kimi model.