Second Brain with AI

The personal knowledge base treated as an agent’s memory substrate rather than a human reading
surface.

The shift

“Second brain” predates AI as a PKM term (Tiago Forte’s framing). What changed through 2025–2026 is
who reads it. The term migrated from note-taking method to agent memory architecture: the
knowledge base stops being where you look things up and becomes what an agent retrieves from and
writes back to.

Timeline

DateFraming
2024-02AutoGen RAG — agents given a second brain
2024-10Agents built to read a personal second brain
2025-03Productised “AI-powered second brain” (Kortex)
2026-08Graphify + Obsidian — “a second brain that never forgets”
2026-08Antigravity + Obsidian as a fully free second brain + agent
2026-08Claude + Obsidian 2.0

The 2026 pattern

Three components recur, and the combination is what’s new:

  1. A plain-text vault — almost always Obsidian, because markdown files are directly readable by agents
  2. An agent with filesystem access — Claude, Hermes, Antigravity
  3. A graph or retrieval layer — Graphify, knowledge graphs, RAG

The economic argument driving adoption is that this stack is free and local: no per-seat SaaS, no
vendor holding the knowledge base.

The pattern is not vendor-specific — what makes it work is plain-text storage the agent can read
and write directly, which is why Obsidian-style markdown vaults dominate over database-backed
tools.