Annotate anything in the browser, hand it to your AI agent, ship. No Slack threads. No tickets nobody reads. No switching context.
Works with any framework · Best results with React · npx @vibeflow-tools/cli kanban · or vf kanban once installed
You review the agent's work in the browser. You spot issues. Then the hard part: re-explaining every little thing in plain text, hoping the agent understands where to look and what's wrong.
One command starts a local server that injects an annotation layer into any HTML file or live app. The rest is just clicking.
localhost:3700 in your browser. Click any element, write a note, set priority. Tasks go straight to .vibeflow/ — versioned in git, live in the Kanban board.
npx @vibeflow-tools/cli tasks --next
.vibeflow/ in your project — committed to git, readable by any tool.Vibeflow ships as two focused packages — use one or both, depending on your workflow.
Vibeflow stores tasks in your project — any tool that can read files or run commands can use them.
Tasks live in .vibeflow/ — plain JSON files in your repo. No API keys. No vendor lock-in. Just structured context your agent can read.
No sign-up. No dashboard. No onboarding. Just install and run.
Requires Node.js 22+. Works with any app, any framework.
Quick answers to common questions.
.vibeflow/tasks/ inside your project — plain JSON files, versioned in git. No database, no cloud service required. You own your data.--next command atomically claims a task (todo → in-progress), so no two agents ever pick the same task. Safe for parallel workflows.