obsidian‑second‑brain

What it is – An open‑source “second brain” that plugs large‑language‑model agents (Claude Code, Grok Bot, Codex, Gemini, etc.) into an Obsidian vault. The agents share a common logbook (called Crew) so they can see each other’s work, and a set of 48 slash‑commands lets you capture, organise, and retrieve knowledge directly from chat sessions.

Why it matters – LLMs normally start each conversation with a blank slate. This project gives them persistent, searchable memory that lives in the files of your Obsidian vault. Your notes are automatically rewritten, contradictions are reconciled, and patterns are synthesised without you having to manually copy‑paste or maintain a separate knowledge base.

Key features (as described in the README)

• Crew logbook – a plain‑folder job record that all agents read/write; no background daemon is required.
• 48 commands organised in four layers:
  1. Operations (31 commands) – saving decisions, ingesting URLs, PDFs, audio, images, exporting the vault, scheduling calendar events, etc.
  2. Thinking tools (9 commands) – challenge your ideas, surface hidden patterns, synthesize insights.
  3. Context engine (1 command) – loads a short “identity” note so each new session starts with your critical facts.
  4. Research toolkit (7 commands) – live X‑post fetch, trend scanning, web research via Perplexity, YouTube transcript + visual analysis, podcast ingestion, etc.
• Auto‑synthesis & scheduled agents – nightly, morning, weekly, and health‑check agents run automatically to reconcile contradictions, create synthesis pages, and rebuild indexes.
• Open Knowledge Metabolism (OKM) – facts are stored with timestamps or pointers, allowing the vault to age out stale information and keep a bi‑temporal record of what was true when.
• AI‑first note format – each note begins with a ## For future agent preamble and front‑matter designed for LLM retrieval rather than human‑only reading.
• Multi‑model support – works with Claude Code, Grok Bot, Codex, Gemini, OpenCode, Hermes, Pi, and others via simple CLI wrappers.
• Zero‑install for Crew – just a curl‑pipe‑bash script; the rest runs on any Python 3.10+ environment without extra packages.

Typical workflow

1. After a meeting, run /obsidian-save; the agent extracts decisions, tasks, people, and creates linked notes.
2. Drop a file or URL with /obsidian-ingest; Claude (or the selected model) transcribes, parses, and rewrites 5‑15 related pages.
3. Use /obsidian-challenge before a decision to have the vault surface past failures.
4. Run /obsidian-visualize to get a graph of your vault’s structure.
5. Nightly, the background agent reconciles contradictions and synthesises new pattern pages.

Installation – One‑line script (curl … | bash) that installs the quick‑install script, then you run /crew setup inside Claude Code (or the equivalent hook for other agents) to point the logbook at a folder inside your vault.

Who it’s for – Developers, founders, writers, researchers, or anyone who already uses Obsidian for note‑taking and wants their LLM assistants to remember and act on that knowledge across sessions.

License & community – MIT‑licensed, actively maintained (v0.18 as of Oct 2026), with a GitHub Discussions forum, a sponsor button, and regular public blog posts by the author.