Summary
Ben Holmes, developer relations lead at Warp, walks through a practical system for turning disorganized personal notes into a navigable, agent-queryable knowledge base. The talk at AI Engineer covers the full pipeline from raw thought capture to enriched wiki, with live demonstrations of each layer.
Holmes starts with voice dictation as the fastest input method, spotlighting two local, on-device tools: Handy (open-source, free) and Voice Inc. (a one-time $20 fee), both capable of transcribing at roughly 200 words per minute with automatic punctuation. He then introduces Hubble, a Markdown-based note viewer he built himself, and demonstrates an agent workflow that enriches raw notes with backlinks, category tags, and cross-references — enabling Wikipedia-style navigation through your own accumulated thoughts. A second agent pass generates structured wikis that group people, concepts, organizations, and sources across an entire note corpus.
The demo spans several personal knowledge domains, including AI news and music-industry content, showing how the enrichment pipeline surfaces connections across notes written weeks apart. Holmes also touches on graph visualizations that are now trivial to generate since agents can write and execute HTML directly. The talk is squarely aimed at developers who want a self-hosted alternative to Obsidian or Notion AI, with full control over the underlying Markdown files and the agent logic that enriches them.
📺 Source: AI Engineer · Published August 12, 2026
🏷️ Format: Tutorial Demo
