Search is the feature people ask for. It is not the interesting layer.
An index that maps keys to glyphs is boring on purpose. Query wraps that index: take a probe, rank candidates, return the neighborhood. The thought process was never “how do we match Elasticsearch.” It was how much work can we refuse to do.
Eliminate work early
Glyph and Atlas optimize around one principle: don’t make the search algorithm do work the system can eliminate earlier. Documents are reduced to compact fingerprints, and multiple fields can be combined into a single searchable representation. Instead of comparing a query against every document, LSH helps narrow the search to a small set of likely matches first, while filters like tags and users let the database eliminate irrelevant documents before similarity scoring.
The result is a deliberate tradeoff: give up some precision and generality in exchange for dramatically lower work per query. Glyph proves the approach in TypeScript. Atlas pushes it further with a lower-level implementation and database optimizations, aiming to keep searches fast even as the dataset grows into the millions of documents.
Building an index is the setup. Query is the opinion: don’t just search. Probe a sketch. Band it. Filter the cheap stuff. Score the remainder. If you need the exact language, use Query to find the right documents, then Spotlight to pull the relevant blurbs/chunks.
Why “related” is the same problem
Recommendations in glyph are not a second model. On this site they are the same index with a different probe; the current post’s fingerprint instead of a user generated string. Once everything is a glyph, ‘search,’ ‘related,’ and ‘more like this’ become the same comparison with different inputs.
That is the payoff of the earlier essays. MinHash keeps the representation small. Collections keep the representation pointed at useful fields. Query keeps the comparison from becoming a giant table scan dressed up with AI.
If you want the APIs, thresholds, and LSH knobs, they live in the Glyph repo and the Glyph-Blog starter. This post is the design note: search is what you ship. Less work per query is what you actually designed (if you followed along!)
I hope this has been a useful resource in getting started with glyph, Thank you for reading <3
Justin 💖