Turn 10,994 Notes Into Memory
summary
A concrete system turns a 10,000+ note Obsidian vault into live agent context: three iterations over 18 months produce a three‑layer pipeline—raw files, an index.yaml catalog, and a wiki layer that updates with queries—implemented as an open‑source AI Research OS repo and daily workflow. It uses Codex/Claude Code and NotebookLM where appropriate, ingests GitHub Readwise Google Drive and YouTube, builds a file‑based index, no DB, creates token‑efficient executive summaries, and exposes a wiki layer that evolves with questions so agents reuse rather than reingest content.
AI-written summary · about
chapters (21)
- 0:00Introduction to the "Second Brain" concept
- 0:49The core problem: Losing research and finding meaningful notes
- 1:32Building an AI Research OS
- 2:05Meet the presenters: Louis-François Bouchard and Pauline
- 3:31Choosing the right tools for research (Google vs. LLMs)
- 4:51Why NotebookLM and vector databases aren't always ideal
- 7:08The need for a personalized research assistant
- 9:16Moving data to Obsidian for local file management
- 11:02Overview of the AI Research OS repository
- 12:48The three-layer system: Raw content, Index, and Wiki
- 13:28Architecture of the Deep Research algorithm
- 18:28Version 3: Adding the Wiki layer on top of knowledge bases
- 20:13How the file-based index works (no database required)
- 21:26Exploring the Wiki structure: Comparisons, concepts, and entities
- 22:49How to query the Wiki efficiently
- 25:01Managing snapshots and personal notes using the PARA method
- 27:04Demo 1: Researching agent engineering
- 31:58Demo 2: Ingesting and comparing GitHub repositories
- 34:25Demo 3: Ingesting custom web links
- 36:50Future improvements: Connectors, memory compaction, and source provenance
- 38:45The Agent Engineering course overview
related talks
12mObsidian
38mWhat the User (Doesn't) Say
23mThe Personal Knowledge Vault
27mThe Future of Notebooks in a Claude Code World
22mHas This Happened Before? Queryable Memory for Database Logs
12mCognitive Exhaust Fumes, or: Read-Only AI Is Underrated
47mRevolutionize Your Note-Taking: From Bullet Journal to Obsidian
01h 09mGive Your Agent a Computer