What this is

Zhira is Shona for “path(s)”. Kindle highlights extraction and organisation were solved by Nota and the CLI; what was left was retrieval - how to find the passage half-remembered, the connection between two books read four years apart, or the common strands in Why Greatness Cannot Be Planned and Obliquity.

The corpus is currently over 14,000 records from 299 books, roughly 620,000 words, going back 6 years.

Zhira is a versioned context pipeline: source data changes, and the system produces a new, checksummed artifact you can diff, verify, and serve, to a person or to a model with citations back to the book and location.

Where it currently is

Early build. What exists today:

• a fresh-ingest pipeline - parse, canonicalise, merge, index, atomic promote;
• sync classification, detecting added, changed, and deleted records against a prior artifact by source checksum;
• a doctor command that validates the manifest and re-verifies recorded checksums;
• BM25 search over an SQLite FTS5 index.

Dense and hybrid retrieval, cited answers, and the MCP server that exposes the corpus to agents are planned.

Github