Recipie
Built an Vercel's agent-browser powered browsing agent that can autonomously navigate, interpret, and act on web pages.
Core trick
Gemini video analysis turns a screen recording into a stored recipe that a browser agent replays on live pages.
Description
Built an Vercel's agent-browser powered browsing agent that can autonomously navigate, interpret, and act on web pages. The experience was smooth Vercel made deployment fast and seamless, and the infrastructure handled real-time interactions reliably. It enabled us to quickly iterate on agent workflows and focus more on product logic rather than setup.
README excerpt
# Recipie Product Requirements Document for **ReplayAgent** — a platform that turns screen recordings into executable browser automation. > *"Record once, automate forever."* ## PRD 📄 **[docs/PRD.md](docs/PRD.md)** — Full Product Requirements Document ## Summary ReplayAgent enables users to: 1. Upload a video (or Loom URL) + context 2. Gemini analyzes the video → produces video artifacts 3. Video artifacts + chat context → stored in database 4. Builder agent derives an executable **recipe** 5. Vercel agents execute the recipe (agent-browser) ## Tech Stack (Planned) - Next.js 15, React 19, TypeScript - Gemini via Google Vertex AI - agent-browser (Vercel Labs) for execution - Database for artifacts + chat context