query
Run DuckDB SQL queries on frontmatter metadata from markdown files, with optional semantic search to find similar documents based on content.
Instructions
Query frontmatter with DuckDB SQL.
Args: glob: Glob pattern relative to base directory (e.g. "atoms/**/*.md"). sql: SQL query string. Reference 'files' table. Columns are frontmatter properties plus 'path'.
Semantic search (when enabled and indexing is complete): - embedding: document embedding vector (NULL if not indexed) - embed('text'): converts text to embedding vector - array_cosine_similarity(a, b): similarity score (0-1)
Example - find similar documents:
SELECT path,
array_cosine_similarity(embedding, embed('search term')) as score
FROM files WHERE embedding IS NOT NULL
ORDER BY score DESC LIMIT 10Returns: Dict with results array, row_count, and columns.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| sql | Yes | ||
| glob | Yes |
Output Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||