Summary
Sharbel A. walks through Nova, an open-source YouTube strategy agent he built to replace the manual idea-validation work that precedes filming. Rather than generating brainstorm lists, Nova follows a strict 10-step research protocol before recommending anything: it checks the creator’s channel history and Notion pipeline, searches YouTube and VidIQ for demand signals, identifies “viral cluster” audience desires, pulls and reads transcripts from high-performing competitor videos, and scores each idea on demand, channel fit, freshness, proof availability, and packaging strength before returning two or three final recommendations with labeled evidence.
The architectural emphasis is on evidence labeling — every recommendation must declare whether it is VidIQ-backed, competitor-backed, or judgment-only — which Sharbel frames as the difference between an idea generator and a strategic advisor. The agent integrates with Notion for pipeline management, requires a logged-in YouTube desktop session for browse access, and is compatible with Claude Code, Hermes agent, Cursor, and Codex for installation.
The repo is available free on GitHub and includes onboarding instructions written for AI coding assistants, so setup involves pasting the link into your preferred agent and asking it to walk through configuration. The video is most useful for solo creators and small teams interested in applying multi-tool agent architectures to content strategy, and for developers looking for a worked example of memory systems, external API integration, and evidence-weighted output design in a practical agentic workflow.
📺 Source: Sharbel A. · Published July 12, 2026
🏷️ Format: Hands On Build
