Querylight TS 0.9.0

Querylight TS 0.9.0

Querylight TS is a pure TypeScript in-memory search toolkit for browser and Node.js applications that need more than fuzzy matching but less than a full search server. Release 0.9.0 is a major step up from 0.2.0: it broadens the query and ranking model, adds structured discovery features such as numeric/date aggregations, strengthens hybrid and semantic retrieval, expands the documentation into a much more complete product surface, and completely overhauls the demo into a richer documentation search application.

Links:

• Demo: https://querylight-ts-demo.pages.dev
• npm package: https://www.npmjs.com/package/@tryformation/querylight-ts
• Repository: https://github.com/formation-res/querylight-ts

Executive Summary

This release is intentionally much bigger than a routine minor version bump.

Compared to 0.2.0, Querylight TS now offers:

• a broader search-engine-style query surface, including best-field, soft-boosting, wildcard, regex, script, rank-feature, and distance-feature querying
• much stronger hybrid retrieval patterns, including vector rescoring on top of lexical candidates
• structured discovery primitives beyond term counts, especially numeric and date aggregations
• clearer positioning as a browser-first search toolkit that combines lexical, semantic, vector, geo, and faceted search patterns in one local package
• a much larger and more cohesive documentation set, including guides, discovery articles, ranking articles, operational guidance, and demo internals
• a substantially redesigned demo that now feels closer to a real documentation search application than a small showcase page

If 0.2.0 established Querylight TS as a capable in-process search library, 0.9.0 makes a stronger case for it as a practical platform for building serious browser and static-site search experiences.

Release Scope

The diff from tag 0.2.0 to this release includes:

• major library feature additions
• new ranking and retrieval options
• new numeric/date indexing and aggregation support
• a large documentation expansion and reorganization
• a full demo overhaul, including semantic search, faceted navigation, richer result views, and better information architecture
• expanded automated test coverage

This release contains roughly:

• 80+ changed files
• nearly 8,000 added lines across code, docs, tests, and demo assets

Library Features By Area

1. Advanced lexical and structured querying

The query model is substantially broader in 0.9.0.

Newly added or now prominently supported query capabilities include:

• MultiMatchQuery for searching across multiple fields as one conceptual text surface
• DisMaxQuery for best-field scoring patterns
• BoostingQuery for soft demotion instead of hard exclusion
• WildcardQuery and RegexpQuery for pattern-driven term lookup
• ScriptQuery for custom JavaScript filtering logic
• ScriptScoreQuery for custom JavaScript scoring logic
• RankFeatureQuery for numeric relevance signals
• DistanceFeatureQuery for recency/closeness-style boosts
• improved BoolQuery behavior and more predictable minimumShouldMatch semantics

Why this matters:

• the library now supports a much more realistic slice of search-engine-style relevance tuning
• callers can express soft preferences, best-field ranking, structured filters, and custom scoring without having to leave the in-memory model
• browser-side search flows can now model more than a simple match-or-no-match search box

2. Discovery and faceting got much stronger

One of the biggest capability jumps in this release is the move from simple term-oriented discovery to broader structured analytics.

New or expanded discovery features include:

• numeric field indexing
• date field indexing
• sum, min, max, avg, stats, and valueCount aggregations
• rangeAggregation(...) for stable human-defined buckets
• histogram(...) for numeric/date distribution views
• dateHistogram(...) for time-oriented exploration
• improved documentation and examples around terms aggregation and significant terms

Practical impact:

• Querylight TS can now support richer faceted navigation and discovery workflows
• result pages can expose counts, distributions, and bounded drill-downs instead of just flat hit lists
• browser apps can build dashboard-like search experiences without introducing a separate analytics backend

3. Hybrid and semantic retrieval matured

0.9.0 makes semantic and hybrid retrieval more usable in practice, not just possible in principle.

The release adds or expands:

• vector rescoring with VectorRescoreQuery
• clearer separation between lexical candidate generation and vector-based reranking
• more documentation around approximate nearest-neighbour search and hybrid search tradeoffs
• better patterns for chunk-based semantic retrieval in docs-style corpora

Why this matters:

• hybrid retrieval no longer has to mean "run vector search over everything first"
• the library can support a more efficient lexical-first, semantic-second flow
• documentation, content, and recommendation experiences can stay local while still benefiting from embeddings

4. Browser-shipped search architecture is better documented and easier to adopt

The project now does a better job of explaining how to use Querylight TS in the environment it was clearly built for: static sites, browser apps, and build-time generated search payloads.

This release strengthens patterns around:

• JSON-serializable index state
• build-time index generation
• browser-side hydration
• shipped documentation corpora
• chunking strategies for semantic retrieval
• performance and memory tradeoffs
• testing search behavior and tuning relevance over time

This is important because the main challenge in client-side search is usually not indexing a toy dataset. It is designing a robust build-time/runtime workflow. 0.9.0 gives users a much more complete path for doing that correctly.

5. Highlighting and result explanation remain part of the core story

Highlighting was introduced earlier, but this release keeps result explanation as a first-class part of the overall product story through improved docs and demo usage.

The practical takeaway is that Querylight TS now presents retrieval and explanation together:

• retrieve ids and scores
• derive snippets and match evidence
• combine search, filtering, and explanation in one browser-shippable flow

That makes the project easier to evaluate as a complete search UI toolkit rather than only a ranking library.

Demo Overhaul

The demo changed dramatically in this release line. It is no longer just a minimal browser example. It has become a richer, more realistic documentation-search application that demonstrates both the library surface and the architectural patterns behind it.

1. "Ask the Docs" semantic search experience

The headline demo addition is a full semantic question-answering style flow for documentation search.

The demo now:

• precomputes documentation embeddings at build time
• splits articles into heading-aware chunks for semantic retrieval
• ships the embedding payload with the static app
• lazily loads the transformer pipeline in the browser
• embeds the user query locally
• matches that query against chunk embeddings to recover semantically relevant answers

This is a meaningful product milestone because it shows that Querylight TS can power a fully local semantic documentation experience without introducing a vector database or hosted search API.

2. Related docs and chunk deeplinks

The semantic work is not limited to question answering.

The demo also now supports:

• related-document discovery
• semantic chunk deeplinks
• tighter mapping between semantic hits and visible documentation sections

That improves both usability and credibility. Instead of showing semantic scores in isolation, the demo turns them into navigable reading paths.

3. Faceted navigation and discovery sidebar

The right-hand sidebar is now a real discovery interface rather than decorative chrome.

It exposes:

• tag facets
• section facets
• API-name facets
• significant terms
• numeric article-length buckets
• compact numeric distribution views

This is important for two reasons:

• it demonstrates that Querylight TS supports filter-heavy exploratory browsing, not just free-text lookup
• it gives users a direct example of how aggregations, range filters, and significant terms work together in an actual UI

4. Better search-state handling and result presentation

The demo’s lexical and hybrid flows have been improved so the app behaves more like a polished search product.

Changes across this release line include:

• improved mode switching
• better Enter-key behavior in Ask the Docs
• a semantic-search busy/loading overlay
• better handling of search state, filters, and result rendering
• richer result and navigation affordances around docs content

The net effect is that the demo now better communicates intended usage patterns instead of feeling like a thin test harness.

5. Build pipeline improvements

The demo build process now does substantially more work for the user:

• it reads and parses the docs corpus
• derives metadata used for lexical indexing and faceting
• computes semantic embeddings
• generates app-ready JSON payloads
• integrates the generated content into Vite builds and dev mode

That makes the demo a useful reference architecture for static documentation search, not just a feature showcase.

Documentation Expansion And Repositioning

The documentation set was heavily expanded and reorganized for this release.

New or substantially improved areas include:

• overview and positioning material
• browser search onboarding
• schema design guidance
• analyzers/tokenization deep dives
• autocomplete and faceted-navigation guides
• relevance tuning guides
• numeric/date indexing and aggregation articles
• wildcard, regex, script, boosting, and dis-max query documentation
• vector rescoring and semantic retrieval patterns
• operational guidance on performance, memory, and testing
• demo-internals writeups explaining how the app works end to end

This is more than a docs quantity increase. It changes how approachable the project is:

• new users get clearer positioning and onboarding
• intermediate users get concrete recipes
• advanced users get lower-level articles on scoring, discovery, and architecture

For a library with this many features, that documentation expansion is itself a release headline.

Developer Experience And Reliability

This release also improves project quality around validation and maintenance.

Notable improvements include:

• expanded automated coverage for advanced query behavior
• dedicated tests for numeric aggregations
• stronger regression coverage around serialization and query semantics
• continued modularization of the internals

That matters because the project is now broad enough that release confidence depends on good behavioral coverage, not just feature breadth.

Why The Version Jumps To 0.9.0

The move from 0.2.0 to 0.9.0 is intentional.

This release represents:

• a substantial expansion of the public capability surface
• a clearer and more mature product story
• a more complete hybrid/semantic/discovery toolkit
• a demo that now showcases an end-to-end application rather than isolated features
• documentation breadth that is much closer to what users expect from a serious library

It is still a pre-1.0 project, but it is no longer at the same stage of maturity as the 0.2.x line.

Upgrade Notes

Users upgrading from 0.2.0 should review:

• the expanded query surface, especially if they want to simplify custom bool/query compositions
• the new numeric/date indexing and aggregation APIs for discovery-heavy UIs
• the semantic and vector rescoring docs if they are building hybrid retrieval
• the reorganized documentation index for updated guidance and examples

If you only used Querylight TS as a simple lexical search library before, 0.9.0 gives you a much larger toolbox without requiring a backend search service.

Validation Checklist For This Release Prep

Before tagging 0.9.0, the repository should still complete the standard release validations:

• npm test
• npm run build --workspace @tryformation/querylight-ts
• npm pack --workspace @tryformation/querylight-ts --dry-run

The release should only be tagged after the release-prep commit is pushed and branch CI is green for that exact commit SHA.