OpenAI-compatible reaches Ollama, OpenRouter, vLLM, a LiteLLM proxy, and OpenAI models up to gpt-4.1/gpt-4o. For a local Ollama model use base URL http://host.docker.internal:11434/v1 and leave the key blank. OpenAI Responses is required for gpt-5 and later, which do not accept tools over chat completions.
http://host.docker.internal:11434/v1
A model marked default answers for every connection that has not pinned one of its own. Pin a database to a local model to keep its schema and rows on this machine.
Running inside a container, localhost means the container itself. Use host.docker.internal to reach a database published on your machine.
localhost
host.docker.internal
Connections are managed by an administrator.
An administrator manages users and connections. That is not the same as access to the data: an admin still has to be granted a connection before they can query it, and granting is a visible act.
Use Access on a connection to choose who may query it. A user sees only the connections they can reach.
Done