dsh-obsidian-agent-wiki

Most file-system memory plugins scan notes one by one in a linear pass. They cannot filter by directory or tag, and—critically—they miss two-character Chinese terms entirely, which are the backbone of a Chinese knowledge base. dsh-obsidian-agent-wiki solves this by indexing your local Obsidian vault into an in-memory SQLite table at startup. Every write or append immediately re-indexes the affected note, so the agent always sees a fresh, searchable snapshot while you continue to browse and edit notes in Obsidian itself, with no frontend required.

The search layer uses SQL LIKE fuzzy matching rather than FTS5 trigrams, a deliberate choice that ensures queries shorter than three characters still return hits. At the scale of tens to hundreds of notes, each scan completes in well under a millisecond. Beyond free-text search, wiki_search supports structured filtering by path, category, project, and tag, returning results ranked by match count along with the note path, title, category, project, and a contextual snippet. Write operations trigger an immediate re-index after the file hits disk, keeping the agent and your vault in lockstep.

If you maintain a Chinese technical wiki or project notes library in Obsidian and want your DeepSeek Harness agent to retrieve and reference that content precisely—by project, by tag, by keyword—without pulling in a single runtime dependency, this plugin was built for you.