Skip to content
Vitrino docs

First run

Vitrino needs two things before it can do anything: a way to pay for the model, and a folder to work in.

Sign in with an API key

Vitrino authenticates with an Anthropic API key. Usage is billed by Anthropic directly to that key; Vitrino has no servers in the loop and never sees your bill.

  1. Create a key in the Anthropic Console.
  2. In Vitrino, open Settings → Account.
  3. Paste the key and save.

Note

Can I use my Claude Pro or Max subscription instead? Not at the moment. Anthropic's Agent SDK terms don't allow a third‑party app to offer claude.ai sign‑in without approval, so released builds of Vitrino offer API‑key authentication only.

Vitrino ignores the key in your environment

If you already use Claude Code in a terminal, you may have ANTHROPIC_API_KEY set in your shell. Vitrino deliberately ignores it, along with every other credential variable Claude Code would otherwise pick up.

Important

An app launched from the Dock or Start menu does not inherit your shell's environment, but one launched from a terminal does — so the same app could bill two different ways depending on how it was opened. Vitrino removes the ambiguity: the credential in Settings is the credential used, always.

Open a project

A session runs in one folder, and Claude can read and write inside it.

Use File → Open Project… (Cmd/Ctrl + O) and choose a folder. Vitrino opens a tab for it and starts a session.

Choose the folder you would cd into before running the agent — usually a repository root. Claude reads the CLAUDE.md files it finds there, exactly as Claude Code would.

Send your first message

Type in the box at the bottom and press Enter. Shift + Enter starts a new line.

Something useful for a first run, because it makes Claude read before it writes:

Have a look around this project and tell me what it does and how it's organised. Don't change anything.

As Claude works, each step appears as its own row — a command, a file read, an edit with a diff. That is the transcript.

When Claude asks permission

The first time Claude wants to run a command or change a file, a permission card appears at the bottom of the transcript and stays there until you answer. It shows exactly what will happen, in full.

Nothing runs until you choose. See Permissions.

Next

The transcript explains how to read a session.