Claude Obsidian: Persistent Wiki for LLM Memory

Claude Obsidian plugin builds a scalable wiki in Obsidian using hot.md summaries, index.md maps, and detailed pages to give Claude persistent memory across sessions, powered by /save, /autoresearch, and /canvas commands with minimal token costs.

Scalable Three-Layer Architecture for LLM Persistence

Adapt Andrej Karpathy's LLM Wiki pattern to create compounding knowledge in Obsidian: load a concise hot.md file (~500 words) every session with quick actions, active research summaries, and key notes; reference index.md for one-line summaries of all wiki pages; pull full wiki pages only as needed for specifics like concepts, sources, decisions, or projects. This keeps context relevant without bloating prompts—wiki scales to thousands of pages while token costs per session remain stable, enabling Claude to recall details from weeks or months prior, like 'best potatoes from two weeks ago,' turning ephemeral chats into a growing second brain.

Obsidian stores everything as plain Markdown files locally, supporting wikilinks, backlinks, graph views, and infinite canvases—used by 1.5 million for knowledge management, now AI-enhanced.

/save Command: Structure Conversations into Wiki

Capture any chat, files, or images with /save: Claude reads the full conversation, generates a dedicated wiki page with YAML frontmatter for metadata, places it in a use-case-specific folder, auto-generates cross-references and backlinks, then updates index.md and hot.md. Result: no more vanishing insights; knowledge integrates with colors, annotations, and interlinks for easy retrieval, compounding like interest across sessions.

/autoresearch and /canvas: Automate Research and Visualization

Run /autoresearch for autonomous deep dives (3-5 iterations): Claude performs broad searches, gap-filling sub-searches on trusted sources, files raw sources as structured pages, synthesizes findings into concept pages with to-dos, then updates index, hot.md, and graphs. Avoids 'tab graveyards' by turning web research into queryable wiki assets.

Use /canvas to generate visual boards: Claude positions flowcharts, text cards, wiki embeds, images, GIFs, or videos in named zones on Obsidian's infinite canvas—ideal for client pitches or audience explainers. Combine with image gen tools (e.g., Nano Banana) by prompting Claude to arrange 20+ generated assets with prompts. Retrieval cascades efficiently: hot.md → index.md → wiki pages, recalling past research instantly.

Two-Line Install for Immediate Use

Install via Claude plugin marketplace: claude plugin marketplace add AgriciDaniel/claude-obsidian then claude plugin install claude-obsidian. Open Claude Code; wiki builds automatically. GitHub: https://github.com/AgriciDaniel/claude-obsidian. Free, open-source for Claude Code (Anthropic's tool).

Video description
Your AI starts from zero every session. Claude Obsidian fixes that. It builds a persistent wiki that grows smarter with every conversation so Claude remembers what you taught it last week, last month, or six months ago. In this video I walk through the 3 commands that make it work: /save, /autoresearch, and /canvas. ⏱ Chapters 0:00 Your AI forgets everything 0:08 What is Claude Obsidian 0:28 How the wiki works 1:06 Command 1: /save 1:52 Command 2: /autoresearch 3:04 Command 3: /canvas 4:40 Under the hood (hot.md, index, wiki pages) 5:43 Install in 2 lines 6:03 What to do next 🔧 Install Claude Obsidian (2 lines) claude plugin marketplace add AgriciDaniel/claude-obsidian claude plugin install claude-obsidian 📦 GitHub Repo https://github.com/AgriciDaniel/claude-obsidian 🧠 What is Claude Obsidian? Claude Obsidian is a free, open-source Claude Code plugin that turns Obsidian into a compounding knowledge base for AI. Based on Andrej Karpathy's LLM Wiki pattern, it gives Claude persistent memory across sessions using a three-layer architecture: hot.md (loaded every session, ~500 words), index.md (a one-line summary of every wiki page), and the wiki pages themselves (concepts, sources, decisions, research). Your wiki can grow to thousands of pages while your token cost per session barely moves. Three commands power it: • /save - files the current conversation into the wiki with proper frontmatter, cross-references, and index updates • /autoresearch - runs an autonomous research loop: searches the web, fetches sources, synthesizes findings, and files everything as structured wiki pages • /canvas - creates visual boards inside Obsidian with flowcharts, images, text cards, and wiki page embeds 💡 What is Obsidian? Obsidian is a free, offline-first note-taking app that stores everything as plain markdown files on your computer. It supports wikilinks, backlinks, graph view, and an infinite canvas. Over 1.5 million people use it for personal knowledge management. Claude Obsidian turns it into an AI-powered second brain. 🔗 Links Claude Obsidian (GitHub): https://github.com/AgriciDaniel/claude-obsidian Claude Code (Anthropic): https://docs.anthropic.com/en/docs/claude-code/overview Obsidian: https://obsidian.md Karpathy's LLM Wiki pattern: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f My website: https://agricidaniel.com 📣 Join the AI Marketing Hub 2,800+ creators, SEOs, and agency owners building with AI tools. Get access to workflows, live Q&As, and every Claude Code skill I build - including claude-seo, claude-blog, claude-ads, and more. Free: https://www.skool.com/ai-marketing-hub Pro: https://www.skool.com/ai-marketing-hub-pro #ClaudeCode #Obsidian #AISecondBrain #ClaudeObsidian #AIMemory #LLMWiki #ClaudeCodePlugin #ClaudeCodeTutorial #ObsidianPlugin #KnowledgeManagement #AITools #Karpathy #AgriciDaniel #AIMarketing

Summarized by x-ai/grok-4.1-fast via openrouter

5658 input / 1468 output tokens in 12393ms

© 2026 Edge