First run
You have the app open and it has nothing in it. This page gets you to a first reply, and points at the rest.
The wizard
Section titled “The wizard”A fresh install opens a four-step wizard: a welcome, your profile, your first server, done. It can be skipped, and everything in it is reachable later from Settings.
The welcome screen also offers restore from a backup, which is the right answer if you are setting up a second machine or replacing one. See Data.
An instance can suppress the wizard entirely with PUBLIC_DISABLE_ONBOARDING=true, for deployments
that arrive pre-configured.
Connecting a provider
Section titled “Connecting a provider”Nothing works until Llooma can reach a model. Settings → Servers → Add a server opens a grid of providers: Ollama, OpenAI, Claude, Infomaniak, or anything OpenAI-compatible.
Pick one, paste an API key if it needs one, and the model list fills in. The details per provider, including where to get a key, are on the providers page.
You can add several. Each connection gets a colour, assigned from the unused ones, and that colour follows its models everywhere they appear, so you always know which endpoint is about to be billed.
In server mode the connections are the admin’s business, not yours: models simply appear in the picker. An admin may or may not allow you to add your own on top.
A first answer
Section titled “A first answer”The home screen is a composer. Type, pick a model from the control in the header, send.
Three things to know from the first message:
- Add context attaches knowledge, images and documents. Each becomes a removable pill.
- The toggles next to the composer arm web search and web fetch for that message only.
- The ring in the composer fills as the conversation does. Grey until 60%, then amber, then red. Tap or hover it for the figures. When it worries you, that is what compaction is for.
The conversation names itself after the first reply.