What it does
- Capture ideas in plain English from the dashboard, from Telegram, or from a Slack /slash command.
- An overnight agent reads the backlog, picks the highest-priority idea, and writes the change.
- The change ships to a staging route on the same site so you can click around the new feature before it goes live.
- Approve and the change merges. Reject and it gets discarded cleanly.
- Every build is logged for review, so you see exactly what changed and why.
How it works
An autonomous coding agent (Claude Code or Codex, depending on the workload) runs the build. It reads the idea, scans the codebase, drafts the change, runs the tests, and serves a preview.
The preview lives behind a staging URL so it's safe to share with stakeholders before approval.
On approve, the code merges and the staging route retires. On reject, everything rolls back without a trace.
The stack
Claude Code (Anthropic) or Codex (OpenAI) as the build agent. Python orchestrator. Git for safe rollback. Flask blueprints for staging routes. SQLite for build state.
Who it's for
Founders and ops leaders who keep saying 'wouldn't it be great if our dashboard could…' and never get the engineering cycles to test it. With this, you do.