What it does Drop a PDF, URL, or text file into your vault and Claude extracts entities, concepts, and cross-references, filing everything as plain Markdown. Ask a question later and it answers from your accumulated notes—citing specific pages, not training data. Between sessions it keeps a “hot cache” so you never have to recap.

The interesting bit Most Obsidian AI plugins are chat windows; this one treats your vault as a living system. It lints for orphans and dead links, flags contradictions with [!contradiction] callouts, and supports multi-writer setups with per-file advisory locks. The hybrid retrieval stack—contextual prefix + BM25 + cosine rerank—comes straight from Anthropic’s September 2024 research.

Key highlights

• 15 Claude Code skills including /autoresearch (3-round web research loop) and /think (10-principle thinking framework)
• First-class methodology modes: LYT, PARA, Zettelkasten, or generic
• Multi-model support: Claude, Gemini, Codex, Cursor, Windsurf
• Companion claude-canvas for visual canvas layouts
• Optional DragonScale Memory extension for log folding and deterministic page addresses
• MIT licensed; v1.9.2 is the current public release

Caveats

• Requires Obsidian plus Claude Code; the Local REST API plugin must run on port 27124 for existing-vault installs
• Some features (early-access builds, Pro community) are gated behind a separate “AI Marketing Hub” membership with its own org mirror
• The README is enthusiastic about capabilities but offers no independent benchmarks or quantitative comparisons beyond the feature matrix

Verdict Worth a look if you’re already living in Obsidian and want AI that writes and maintains your notes, not just answers questions about them. Skip it if you need a turnkey SaaS or don’t want to orchestrate Claude Code + Obsidian + plugins.