Descriptions:
Matthew Berman showcases four open-source GitHub projects he considers immediately useful for AI practitioners, walking through live demos of each. The first is Last 30 Days (40,000+ GitHub stars), created by Matt Van Horn — a Lyft co-founder — which functions as a trending-content search engine that aggregates Reddit, Hacker News, X, YouTube, TikTok, and Polymarket results ranked by human engagement rather than algorithmic ranking. It installs as a skill in agentic platforms like Claude Code and outputs shareable HTML briefs.
The second is a fully local, open-source clone of Google’s NotebookLM, allowing users to upload PDFs and URLs, ask questions against them, generate multi-host AI podcasts (up to 23+ minutes, with customizable tone and script editing), and run document transformations including summary, insight extraction, and reflection questions. It supports models from GPT-5.5 and GPT-4o down to fully local Ollama and LM Studio deployments.
The third is Agent Skills (56,000+ GitHub stars), a curated skill library for agentic engineering workflows. The fourth project — teased as capable of cutting AI API costs by up to 90% — rounds out a fast-moving set of tools aimed at developers extending their LLM and agentic pipelines. Berman also flags a growing security concern: malicious code is increasingly being embedded in third-party skill hubs, making source vetting essential before installing community-contributed skills.
📺 Source: Matthew Berman · Published June 12, 2026
🏷️ Format: Roundup
