Configure a provider

Configure at least one provider, then choose one of its enabled models from the provider/model selector in the chat composer.

Connection methods

Settings page	Use it when	Credential or prerequisite
ChatGPT Plus/Pro	You want to use an OpenAI ChatGPT account	Browser login or device code
Models.dev	Your provider is in the models.dev catalog, or exposes an OpenAI-compatible API	Provider API key
Kora	You want to use Flash with Kora Premium	Kora key
ByteDance (ModelArk)	You use the Standard API or Coding Plan	ModelArk API key
Kimi Code	You have a Kimi API key for the Kimi coding models	Kimi API key
OpenCode CLI	OpenCode is installed and already authenticated	Working opencode executable
Devin CLI (ACP)	Devin CLI is installed and already authenticated	Working devin executable
Local	Ollama, LM Studio, omlx, or another local OpenAI-compatible server is running	Local server URL; API key only if required

ChatGPT Plus/Pro

1. Open Settings → ChatGPT Plus/Pro.
2. Select Sign in with OpenAI.
3. Complete authorization in the browser.
4. Return to Kora and select Refresh Models.
5. Check the account-specific list under Available models.
6. Close Settings and choose one of those models in the composer.

If browser authorization cannot return to Kora, select Or sign in with a device code, open the displayed verification URL, and enter the code. Use Disconnect to remove the account from Kora.

The available list is the catalog returned for that account. Do not assume that every model associated with an OpenAI plan is supported or enabled in Kora.

Models.dev

1. Open Settings → Models.dev.
2. Search Search providers from models.dev….
3. Select the provider.
4. Confirm the prefilled Base URL and enter its API Key.
5. Review the model list and select Add Provider.
6. Open the added provider to enable or disable individual models, edit their capabilities, add a custom model, or refresh the catalog entry.

Use Add manually when the provider is absent from the catalog. See Add providers with Models.dev.

Flash with Kora Premium

1. Open Settings → Kora.
2. Paste the Kora key in API Key.
3. Wait for the validation state, then select Save.
4. Close Settings and select Flash in the composer.

See Connect Kora Premium for account and key creation.

ByteDance ModelArk

1. Open Settings → ByteDance (ModelArk).
2. Paste the API key.
3. Under API Mode, choose Standard API (Seed Models) or Coding Plan (Subscription).
4. Leave Custom Base URL empty to use the endpoint for the selected mode, then select Save.
5. Enable the required models in the model manager.

Kimi Code

1. Open Settings → Kimi Code.
2. Paste the Kimi API key and select Save.
3. Enable the required models in the model manager.
4. Close Settings and choose a Kimi model in the composer.

Kimi models keep a minimum generation temperature of 1.0; Kora normalizes lower values automatically. See Add a custom provider.

OpenCode CLI

1. Install and authenticate OpenCode outside Kora.
2. Open Settings → OpenCode CLI.
3. Confirm that Kora detects the executable and loads the models returned by OpenCode.
4. Select an OpenCode model in the composer.

Credentials remain managed by OpenCode. Kora does not ask for an OpenCode API key.

Devin CLI

1. Install and authenticate Devin outside Kora (devin auth login).
2. Open Settings → Devin CLI.
3. Confirm that Kora detects the executable and loads the model catalog.
4. Select a Devin model in the composer.

Kora runs Devin through the Agent Client Protocol (ACP), so Devin can keep its native session and command model inside the Kora workspace. Credentials remain managed by the Devin CLI: Kora does not ask for a Devin API key. Devin’s slash commands appear in the composer palette while a Devin model is selected.

Local

1. Start the local inference server.
2. Open Settings → Local.
3. Select a preset or enter a local OpenAI-compatible base URL.
4. Use the form’s server discovery action to load models, then add the provider.
5. Enable the models that should appear in the composer.

Troubleshooting

• ChatGPT shows no models → select Refresh Models; if the account still returns none, disconnect and authenticate again.
• A Models.dev provider fails its connection test → verify its base URL and API key; catalog metadata does not supply your credential.
• OpenCode shows no models → run opencode models in a terminal and confirm that the executable is available on Kora’s PATH.
• A local server is not discovered → start the server first, verify its port, and use a loopback URL such as http://127.0.0.1:<port>/v1.