claude-obsidian

Product Introduction

1. Definition: claude-obsidian is a free, open-source (MIT-licensed) Claude Code plugin that functions as an autonomous knowledge management agent for Obsidian. Technically, it is an implementation of Andrej Karpathy's LLM Wiki pattern, automating the creation and maintenance of a connected knowledge graph from raw source materials.
2. Core Value Proposition: It exists to solve the critical maintenance problem inherent in traditional "second brain" or Personal Knowledge Management (PKM) systems. It automates the labor-intensive tasks of reading sources, writing structured notes, creating bidirectional links, and maintaining a living index, transforming Obsidian into a self-organizing AI second brain that compounds in value without manual upkeep.

Main Features

1. Autonomous Wiki Ingestion & Compilation: The plugin's core function is the /wiki-ingest skill. It processes raw sources (PDFs, URLs, transcripts) placed in a designated folder. Using the Claude API, it reads the content, extracts named entities (people, companies, tools) and key concepts, and then autonomously writes or updates plain Markdown wiki pages. It creates entity pages, concept pages, and cross-references them against the entire existing vault, flagging contradictions and building a dense network of links automatically.
2. 15-Skill Knowledge Operating System: Beyond ingestion, claude-obsidian v1.9 includes 15 distinct skills that cover the full knowledge lifecycle. Key skills include: /wiki to scaffold a new vault; /save to file an entire AI conversation as a clean, linked note; /autoresearch to autonomously research a topic, read sources, and write up to 12 linked pages; /canvas for building visual reference boards; and /think which runs a 10-principle reasoning loop for complex decision-making.
3. Compound Vault Architecture with Hybrid Retrieval: The v1.9 "Compound Vault" update introduces a sophisticated, multi-layered retrieval system. Instead of relying solely on vector similarity, it combines a contextual prefix, BM25 keyword search, and cosine reranking (based on Anthropic's contextual retrieval research) to pull the most relevant pages from the vault accurately, reducing retrieval failures. The architecture is file-based, consisting of a hot cache, an index, and wiki pages, with no external vector database or embedding server required.

Problems Solved

1. Pain Point: The "maintenance tax" and subsequent abandonment of traditional PKM systems. Users collect notes but fail to create meaningful links, leading to disconnected "note graveyards" where information becomes unfindable. Manual linking and filing are unsustainable bottlenecks.
2. Target Audience: Knowledge workers, academic researchers, content creators, and lifelong learners who manage high volumes of information. Specifically, individuals who have tried and failed to maintain systems like Zettelkasten, PARA, or LYT due to the manual overhead, and those seeking a locally-owned alternative to cloud-based AI memory features.
3. Use Cases: Automatically building a literature review wiki from a folder of academic PDFs; maintaining a constantly updated competitive intelligence vault from news articles and reports; creating a linked knowledge base from meeting transcripts and internal documents; conducting hands-off preliminary research on a new topic where the AI reads sources and synthesizes connected notes.

Unique Advantages

1. Differentiation: Unlike cloud-based note-taking apps with AI features (e.g., Notion), claude-obsidian ensures complete data ownership by storing everything as plain Markdown files on the user's local disk. Unlike other AI-PKM tools that rely on vector databases and background workers, it uses the simpler, more transparent LLM Wiki pattern, eliminating hosting costs, background RAM usage, and complex infrastructure.
2. Key Innovation: The implementation of a "hot cache" – a single plain-text file that stores ~500 words of recent context. This solves the LLM "session amnesia" problem by pre-loading the AI's working memory at the start of a new conversation, saving significant tokens and time otherwise spent re-establishing context. This, combined with its methodology-agnostic ingestion and multi-writer file locking, creates a uniquely resilient and efficient autonomous knowledge system.

Frequently Asked Questions (FAQ)

1. How does claude-obsidian compare to using a vector database for a second brain? claude-obsidian bypasses vector databases entirely by using the LLM Wiki pattern, where the LLM directly reads and writes human-readable Markdown files. This eliminates the need for an embeddings server, reduces costs and complexity, and ensures your knowledge base is always directly accessible and portable without any specialized software.
2. Is claude-obsidian completely free to use? The claude-obsidian plugin itself is free and open-source under the MIT license. However, you must pay for your own usage of the Claude API (or another compatible LLM API) for the AI processing. There is no subscription fee for the plugin, and Obsidian is free for personal use, aligning with a "own your data, rent the intelligence" model.
3. Can multiple people collaborate on a single claude-obsidian vault? The v1.9 "Compound Vault" update introduced per-file advisory locking for multi-writer safety, allowing several AI agents to ingest sources simultaneously without corrupting the vault. However, it is primarily designed for individual knowledge management; for full real-time human collaboration, cloud-based platforms may be more suitable, albeit with less data ownership.
4. What happens to my knowledge base if I stop using the AI or the Claude API? Your entire knowledge base remains fully intact as a folder of standard Markdown files. You can continue to browse, search, and edit it in Obsidian or any text editor indefinitely. The AI is a tool that builds and maintains the structure, but the output is permanently yours in a future-proof format.
5. How does the autonomous ingestion handle inaccuracies or contradictions in sources? The /wiki-ingest skill is designed to flag contradictions between new information and existing notes in the vault, alerting the user. However, the accuracy of the generated content ultimately depends on the source material and the capabilities of the underlying LLM (Claude). The system operates best with credible sources and benefits from user review, functioning as a powerful augmentation tool rather than a fully autonomous, infallible agent.