Model connections
@create openai
@openai@create openai asks for a name, a provider, an endpoint, a model or deployment, an API key, and an optional system prompt.
| Provider | What you fill in |
|---|---|
| Azure OpenAI | The resource endpoint, the deployment name, and the key. This is the default. Existing connections stay Azure |
| OpenAI or compatible | A base URL (blank means https://api.openai.com/v1), a model, and a key sent as a Bearer token. The key is optional for a local server |
Ollama, LM Studio, vLLM, and a gateway work as the compatible base URL. Both Azure URL styles are accepted: an endpoint that contains /openai/v1, and a classic deployment host.
The system prompt is stored on the connection. It is not a secret. Blank or whitespace sends no system message. Test is a small chat completion and does not send the system prompt. Save stays disabled unless the deployment accepts the key. Changing only the system prompt does not require another test.
The key goes to the OS keychain, or you type a vault reference instead. Secrets covers akv:// and op://.
Select it for this tab
@openai selects one connection for the current tab. It does not edit. The status strip shows the name as a pill. Hover that pill and confirm Disconnect? to clear it for this tab. A saved prompt is picked up on the next interpretation without selecting the connection again.
termorq-cli models lists these connections from the app data folder. It does not call them. A collection run skips @chat, @agent, and / skills.
termorq-cli models --data-dir ~/Library/Application\ Support/com.garrardkitchen.ai-terminal--workspace is an error here. OpenAI connections are in the data folder.