Browse / LLM wiki / Second brain / boiling-brain

boiling-brain

tetra-plg

Opinionated implementation of Karpathy's LLM Wiki pattern — maintained by domain-expert agents (Claude Code).

Open tetra-plg/boiling-brain → Checked against the four rules.

agent-basedclaude-codekarpathyllmllm-wikiobsidianpersonal-knowledge-basepkmsecond-brainwiki-template

Built for Claude Code

ships .claude and nothing for other agents.

In its own words

README.mdtetra-plg/boiling-brain

Bootstrap template for an LLM Wiki — a personal knowledge base curated by you and maintained by LLM agents.

Status

v1.1.0 — major release: an MCP server for cross-project wiki access (tiered loading, ~96% token reduction), a session lifecycle (Stop/SessionStart hooks + /compress-bb), /domain lifecycle commands, L3 readiness (ADR verdict tracking), and a CI revamp for living vaults (Obsidian-safe Prettier formatter, semantic-only linting, deterministic wiki validation). The template works end-to-end and has scaffolded real vaults. See CHANGELOG.md for the full milestone list. Bug reports and generic-improvement PRs welcome — see CONTRIBUTING.md.

What is an LLM Wiki?

You curate sources (notes, transcripts, articles, repo docs) into an immutable raw/ folder.
Domain expert agents ingest those sources and write the wiki layer (wiki/sources/, wiki/concepts/, wiki/entities/, etc.) — one agent per domain you care about.
Slash commands (/ingest, /query, /lint, /evolve-agent, …) orchestrate the agents.
The wiki is **always derivable from raw/** — no orphan knowledge, no hallucinated links.

The pattern follows Karpathy's "knowledge compilation" idea: keep the source of truth compact, hash-addressed and immutable; let LLMs maintain a queryable layer on top.

Quoting 7% of the README. Read the rest on GitHub ↗

Read from the repository's git tree on 2026-08-17.