Connect a provider and run your first task¶
Ru has no account of its own. You choose a model provider, give it your own API key, open a project and ask for a change.
Connect a provider¶
In the desktop app, open Settings → Models & Providers:
- Pick the Active Provider.
- Paste your API key. It is stored in Windows Credential Manager and sent only to the provider you chose.
- Select Test connection to confirm Ru can reach the provider.
- Close Settings and pick a model beside the message box.
In the terminal UI, use slash commands (type / to see them):
/backend choose the provider
/key <provider> enter its API key (masked; never printed or logged)
/model choose a model from the live catalog
/reasoning set the reasoning effort, if the model offers levels
Providers are OpenAI, Anthropic, a Custom OpenAI-compatible server (a keyless endpoint works), and every
provider in the models.dev catalog, listed as models.dev:<provider-id>. Models
the catalog marks as free carry a free tag in the terminal picker.
Note
Ru never switches provider, model, key or reasoning effort on its own. If a request fails, the run reports it and keeps the selection you made.
Open a project¶
- Desktop: open your project folder from the sidebar.
- Terminal: run
ruinside the folder, or start a chat for any folder withru agent chat --workspace PATH.
The first time Ru sees a folder it asks you to trust it. Nothing runs until you decide. Trust is tied to that folder's identity, so a moved or recreated folder asks again.
Ask for something small and read-only¶
Start with a task that cannot change anything:
Find where this project handles configuration. Explain it before changing anything.
For a first run, choose the plan permission mode, which allows reading and searching but denies edits
and terminal commands. In the desktop app, find it under Settings → General → Permission Preset; in
the terminal, start with ru agent chat --preset plan. Permissions explains all three
modes.
Follow the work¶
Ru shows what it does as it does it.
- Desktop: an activity feed lists each tool call, a changes view shows edited files, and an integrated terminal runs beside the chat.
- Terminal: each tool call is a row with its target and result inline. Press
Ctrl+Oto expand the model's reasoning. PressEscto interrupt a turn.
When Ru needs your permission it stops and asks. Read the command or path in the request before you approve it.