Connecting a model provider
Connecting a model should not force you to remember whether Anthropic uses /v1, whether DeepSeek is
OpenAI-compatible or which port Ollama opens. Under Connectors → Models → Provider there is a row
of shortcuts: press one and the name, the type and the base address fill themselves in, leaving only
the key to paste.
The ones that come set up
Section titled “The ones that come set up”| Provider | Type | Address |
|---|---|---|
| OpenAI | openai |
the client’s own |
| Anthropic | anthropic |
the client’s own |
| Google Gemini | google |
the client’s own |
| DeepSeek | deepseek |
https://api.deepseek.com/v1 |
| Ollama Cloud | ollama |
https://ollama.com/v1 |
| Ollama (local) | ollama |
http://localhost:11434/v1 |
| OpenRouter | openrouter |
https://openrouter.ai/api/v1 |
| Groq | groq |
https://api.groq.com/openai/v1 |
| Mistral | openai_compatible |
https://api.mistral.ai/v1 |
| xAI (Grok) | openai_compatible |
https://api.x.ai/v1 |
| Together AI | together |
https://api.together.xyz/v1 |
| Azure OpenAI | azure |
your resource’s own |
| vLLM (your own) | vllm |
http://localhost:8000/v1 |
| LM Studio (local) | lmstudio |
http://localhost:1234/v1 |
The models are not in the code
Section titled “The models are not in the code”A reference list frozen in code goes stale in weeks: the provider ships new models, retires old ones
and moves prices. That is why the catalogue only stores the provider, and the models are
discovered by asking it with «Show models», which calls GET /models with your key and lists
them. You add them with one click.
Whatever arrives that way comes in at zero price, because the provider does not publish what it charges you. That zero does not mean «free»: the euro cap is not applied until the rate is set, and the role record says so.
If a provider has no base address written down (OpenAI, Anthropic, Google and Azure carry it inside their client), discovery warns you rather than inventing a URL.