jessepinkman9900/claude-second-brain
Productivity & UtilitiesScaffolds a Claude Code–maintained Obsidian wiki with local hybrid search and Git sync for personal knowledge management.
What it is
claude-second-brain is a CLI that scaffolds a personal knowledge base where Claude Code ingests sources, writes cross-linked wiki pages, and answers questions with citations. The scaffolded folder is itself a valid Obsidian vault, and qmd provides local hybrid search (vector plus BM25) over the wiki. GitHub acts as the versioned source of truth.
The design follows an "idea file" approach: CLAUDE.md stores the schema Claude reads each session, defining five page types (overview, topic, entity, source-summary, qa) linked with Obsidian wikilinks and flagged contradictions. It is not a RAG chatbot over notes; Claude actively maintains a structured wiki.
Quick start
Requirements: Node.js, Claude Code, and optionally Obsidian. The CLI can install mise and the Node/pnpm toolchain if missing.
- Scaffold a brain:
npx claude-second-brain
Or install globally and use the csb alias:
npm i -g claude-second-brain
csb
The prompt asks for a brain name (default my-brain) and optionally a GitHub repo, which it creates and pushes to if the gh CLI is available. It creates the brain under ~/.claude-second-brain/my-brain/, registers it in config.toml, installs dependencies, and runs git init.
- Build the search index:
cd ~/.claude-second-brain/my-brain
pnpm qmd:reindex
The first run downloads roughly 2 GB of GGUF models, once per machine.
- Open
~/.claude-second-brain/my-brain/as a vault in Obsidian. The bundled.obsidianfolder preconfigures the obsidian-git plugin; enabling it syncs automatically. If no GitHub remote was created during setup, add one manually withgit remote add origin <url>and push.
Then drop sources into raw-sources/articles/, raw-sources/pdfs/, or raw-sources/personal/, run /brain-ingest in Claude Code, and use /brain-search for cited answers.
Skills and multiple brains
The workflow is driven by slash commands: /brain-ingest extracts knowledge and cross-links it, /brain-search runs hybrid search and can file answers as wiki/qa/ pages, and /lint surfaces orphan pages, broken links, and unresolved contradictions. Maintenance commands include /brain-refresh for re-embedding changed files and the destructive /brain-rebuild for redesigning the qmd schema.
The three core skills install globally to ~/.claude/skills/; /brain-rebuild and /lint live inside the vault. To update template skills, the README points to npx skills add against the template skill directory. Multiple brains can coexist, each with its own folder, index, and remote; global skills always operate on the default brain declared in config.toml, managed with commands such as csb ls, csb use, and csb doctor.
Constraints
- Claude Code is required for ingestion and search workflows; the CLI only scaffolds and manages brains.
- The first index build requires downloading about 2 GB of local models.
- Mobile editing depends on Claude Code's GitHub integration, and iOS reading relies on placing the vault in iCloud Drive for Obsidian Mobile.
/brain-rebuildis explicitly destructive: it drops the old index and rebuilds embeddings.- Licensing is MIT.
HotRepo's take
The project recorded 14 new stars on the JavaScript daily list on 2026-11-16, reaching 11th place with 166 stars and 23 forks overall. That single daily record is the only available ranking datapoint, so it supports a one-day observation rather than a trend. The repository is young (created 2026-04-11), and its appeal rests on wiring together existing tools — Claude Code, qmd, Obsidian, and GitHub — into one scaffolded workflow rather than implementing search or LLM capabilities itself. Readers already using Claude Code and Obsidian are the natural fit; others inherit the 2 GB model download and a multi-tool dependency chain.
Written by AI from the project's README, updated . The project's own docs are the source of truth.
You might also like
- eugeniughelbur/obsidian-second-brainGives AI coding agents a shared, plain-markdown memory in your Obsidian vault across sessions and tools.
- huytieu/COG-second-brainTurn a local Markdown vault into an AI-maintained second brain with skills, agents, and a verification loop.
- Timaga/noda-local-ai-notesWindows local-first Markdown note app with knowledge graph, semantic search and cited-source chat.
- TriliumNext/TriliumA self-hosted, local-first notes app for people building a deeply linked personal knowledge base that they control.
- avenoxai/avenoxbeyinLocal Markdown memory system that lets AI coding agents save, retrieve and reuse notes across sessions.