Claude Code is brilliant. It just can't work while you sleep.
A CLI is one session, one terminal, one person watching it. VB Box is the board that runs that same CLI — Claude Code, Codex, OpenCode or Gemini — one card at a time, each in its own git worktree, each reviewed by a second agent and tested by a third, each with a spend ceiling. You are not replacing your coding agent. You are giving it a queue, a review gate and a team.
Not an alternative to Claude Code. The board that runs it.
What the board adds to the CLI you already use
It runs your CLI, it isn't a clone of one
VB Box shells out to the real thing — `claude -p`, `codex exec`, `opencode run`, `gemini` — and resumes the same session by id. Pick the executor per card: a cheap model for mechanical work, a frontier model for the hard card.
Your key, your tokens, no markup
You bring your own provider key, so inference is at list price and we never take a margin on it. The $30/seat buys the board — the queue, the worktrees, the review and test passes, the audit log — not resold tokens.
It stops before it overspends
A per-task and per-day budget is checked before a turn is allowed to start. A run that would exceed it is halted and reported instead of quietly burning overnight — which is the thing that stops people leaving a CLI running in the first place.
One terminal doesn't scale to a team
A CLI session lives on one laptop and dies with it. Cards are shared: roles decide who can run what, several agents work the same repo in isolated worktrees, every run is in the audit log, and a PM can file the card a developer never has to translate.
Cost you can attribute, not just a bill
Spend is tracked per card, alongside an economy score, continuation share and cache reuse. You can see what a feature cost — not just what the month cost. It's a panel on your own board, not a promise on ours.
See what it would build — free
Describe the work and get real wireframes, an architecture sketch and a plan broken into cards, before you pay anything.
Frequently asked questions
Do I still need my own Claude subscription or API key?
Yes, and that is deliberate. You bring your own key, so you pay the provider directly at list price and we take no margin on inference. VB Box charges $30 per seat per month for the board itself.
Is this just a wrapper around Claude Code?
It runs Claude Code, but the product is everything around it: a queue, git-worktree isolation per card, a second agent that reviews the diff, a third that tests it, a human approval gate on the design, per-task budgets, roles and an audit log. The CLI is the executor, not the system.
Can it use models other than Claude?
Yes. Claude Code, Codex, OpenCode and Gemini are all supported executors, chosen per card, so you can run cheap models on mechanical work and reserve the expensive ones for the cards that need them.
Is it actually cheaper than running the CLI myself?
Not necessarily on raw tokens — a board that reviews and tests every change does more work than one unsupervised session. What it changes is the cost per shipped, reviewed and tested change, the absence of any markup on your inference, and the fact that spend is capped before a turn starts rather than discovered on the invoice.
What happens when an agent gets it wrong?
The card fails its review or test pass and goes back, rather than merging. Design is approved by a human before implementation starts, every run is in the audit log, and each card is isolated in its own git worktree, so a bad run doesn't touch anyone else's work.
Can non-developers use it?
That is the point of a board. Anyone on the team can file a card in plain language; the agent picks it up, and a human still approves the design and the merge.