Quickstart

A server to talk to

Everything below needs a deployment: a workspace on app.vitrinka.ai, or your own host — installing on your own host is one Compose file. The CLI and the recorders work identically against both.

Get an account

No account yet? Sign up at app.vitrinka.ai/auth/signup — the beta is open, the account is created on the spot, and the login page links there too. On your own host, your first sign-in creates your account, its organization and a first workspace.

Install and set up

From the repository you are working in:

npm i -g @vitrinka/cli
vitrinka setup

One install for macOS, Linux and Windows. vitrinka setup signs this machine in through a browser approval — never a pasted secret — and keeps the write token that sign-in issues, installs the Claude Code / Codex plugin with its skills from the public vitrinka-kit marketplace, registers MCP (a secret-free HTTP entry that signs in with OAuth), binds this checkout to its project in a committed vitrinka.config.json, adds the agent hooks and, if you accept, the browser recorder extension. Flags and what is prompted for: installing the CLI.

Pick a workspace

Belong to more than one workspace? Bind this repository to the one its writes should land in:

vitrinka config bind acme --project myapp

The bind tells every future command — and every agent session in this repo — which workspace and project it belongs to.

Put something on a board

The fastest first board is an upload:

vitrinka board init                            # once per repo — names the set and prints its root
vitrinka board capture web --file shot.png --route /home --note "Home"  # adopt a screenshot into the set
vitrinka board push                            # publish the set to the server
vitrinka board create --title "First board"  # imports the set — prints the board URL

More files later go straight to the board: vitrinka board upload ./next.png --board <slug>.

Or let your agent do it: say "publish this" while it is building UI and the capture skills screenshot the work and hand back a board URL. Recording a whole manual-testing journey is the browser recorder.

Annotate it

Open the board URL. Draw a region on the thing that is wrong and write one sentence. That is the entire report — the screen, the region and the branch are already attached.

Watch it get fixed

In the app's repository, an agent session runs /vitrinka:listen. Each annotation wakes it: it fixes the code, hangs the after-shot beside your region, and replies in the thread. You accept the fix on the board.

Where next

• SaaS or self-hosted — which deployment fits.
• QA & Testing — recording instead of reporting.
• Developer — the CLI, MCP and skills in depth.