Browse / Second brain / second-brain-starter
second-brain-starter
tonyazhuuki
Minimal Second Brain template: structure + local setup + data tools
Open tonyazhuuki/second-brain-starter → Checked against the four rules.
Works with any agent
ships config for claude, cursor.
In its own words
README.mdtonyazhuuki/second-brain-starter
A ready-to-use knowledge base template for Claude Code + Obsidian, based on Karpathy's LLM Wiki architecture.
Quick Start
- Fork/clone this repo
- Open in Claude Code:
cd second-brain-starter && claude - Ingest your first source: `/ingest
- Claude Code will create raw transcript + synthesis + update index
Architecture (Karpathy LLM Wiki)
- Raw (
01_library/raw/) — immutable source material. Never edited. - Wiki (rest of
01_library/) — LLM-generated synthesis. Two page types: - Source note — 1 source = 1 file (what author X said about topic Y)
- Topic page — 1 topic = compiled knowledge from ALL sources
- Schema (
CLAUDE.md+.claude/rules/) — rules for the AI
How Ingest Works
One ingest should touch 5-15 files (not just create one isolated file).
When Topic Pages Are Created
After 3+ source notes on the same topic, create a topic page in topics/ that compiles all knowledge into one document. Reading a topic page = understanding the topic.
Customization
- **Edit
CLAUDE.md** — add your communication preferences, project context - **Edit
.claude/rules/** — add domain-specific rules - **Add
.claude/commands/** — create custom skills
Works With
- Claude Code (primary) — terminal AI assistant
- Obsidian (optional) — visual navigation, graph view
- Git — version control for your knowledge
Credits
- Architecture: Andrej Karpathy — LLM Wiki
- Template: @tonyazhuuki
The whole README. Open the repo ↗
Read from the repository's git tree on 2026-08-17.