AgriciDaniel/claude-obsidian

What it solves

claude-obsidian transforms a standard Obsidian vault into a local-first, provenance-aware knowledge system. It prevents the common AI note-taking problem where summaries are disconnected from their sources, ensuring that every claim made by an AI agent is grounded in evidence and traceable back to original documents.

How it works

The system operates as a plugin for Claude Code and other Agent Skills compatible hosts. It manages a directory of Markdown, JSON, and source files using a repeatable loop: capturing local sources into an immutable inbox, synthesizing them into linked notes, and using those notes to answer queries.

To ensure data integrity, it uses a transaction-based write system: workers generate drafts, an orchestrator bundles them into a plan, and the user must explicitly approve a SHA-256 hash of that plan before any files are modified. It supports various organizational methodologies like Zettelkasten, PARA, and LYT.

Who it’s for

Researchers, knowledge workers, and Obsidian users who want to leverage LLMs to organize and synthesize information without sacrificing ownership of their files or losing the link between a summary and its source evidence.

Highlights

• Local-first architecture: Your vault remains a set of ordinary files on your disk, not locked in a proprietary database.
• Provenance tracking: Maintains source and claim ledgers to track authority, confidence, and contradictions.
• Transaction-safe writes: Prevents silent overwrites and data loss through a mandatory review-and-apply workflow for all mutations.
• Integrated skill set: Includes specialized tools for web research (autoresearch), vault cleaning (wiki-lint), and visual mapping (canvas).
• Methodology flexibility: Supports multiple filing conventions (Generic, LYT, PARA, Zettelkasten) for routing new notes.