Thirty years of mature libraries, native to your app

Your Kotlin code sits on top of the JVM — the most professionally engineered runtime ever built. AI, search, document parsing, scientific computing, RPC at scale: every problem already has a hardened library on Maven Central. No FFI, no bridges, no JS shim. Same process. Same memory. Already there.

TL;DR

• The JVM library ecosystem ships in your binary — no IPC, no out-of-process daemon.
• All six pillars (AI/ML, search, data, documents, imaging, networking) have mature options.
• ~500,000 artifacts on Maven Central — anything you'd reach for is one implementation away.
• Two Nucleus modules wrap common desktop UX gaps (file dialog, spell check) by pointing at the best existing libraries.

AI & ML

Run inference locally, call remote models, embed an LLM in your app — the JVM has bindings for every framework that matters.

• DJL — Deep Java Library, multi-engine model serving.
• ONNX Runtime Java — production ONNX inference.
• LangChain4j — agent/RAG/tool-use scaffolding.
• llama.cpp via FFM bindings — local LLM inference.
• TensorFlow Java — Google's framework, JVM port.

Search & indexing

The world's most deployed search stack runs on the JVM. Lucene is the engine behind Elasticsearch, Solr, OpenSearch — embed it directly.

• Apache Lucene — embedded full-text search.
• OpenSearch client, Elasticsearch client — talk to clusters.
• Tantivy via JNI — Rust full-text engine.
• Quickwit — cloud-native search.

Data & analytics

Columnar, in-process, distributed — pick your engine. The JVM is where serious data work happens, and now it ships in your desktop binary.

• Apache Arrow — columnar in-memory format.
• DuckDB JDBC — embedded OLAP.
• Apache Parquet — file format on disk.
• jOOQ — type-safe SQL builder.
• Apache Spark — in-process for medium data.

Documents & parsing

Parse anything users drag into your app. PDFs, Word, Excel, raw HTML, 1000+ formats — without shelling out to native binaries.

• Apache Tika — 1000+ formats.
• Apache PDFBox — PDF read/write.
• Apache POI — Office documents.
• Jsoup — HTML parsing.
• commonmark-java — Markdown.

Imaging & vision

Production-grade image processing and computer vision — used by NASA, BioMed, defense. Now it's a Gradle dep away from your UI.

• BoofCV — computer vision in Java.
• JavaCV — OpenCV bindings.
• TwelveMonkeys — 40+ image formats.
• Skia — GPU 2D (already in your Compose process).
• FFmpeg via JavaCV — video.

Networking & RPC

Battle-tested at the scale of Twitter, Apple, LinkedIn. Netty handles billions of connections daily — runs in your desktop app too.

• Netty — async I/O at scale.
• Ktor — Kotlin-native HTTP client/server.
• gRPC Java — RPC framework.
• OkHttp — HTTP client.
• WebRTC Java — real-time media.

For HTTP in particular, Nucleus ships native-ssl and native-http variants that wire the OS-managed trust store into any client (java.net.http, OkHttp, Ktor). Corporate proxies and user-installed CAs just work.

The half-million

Every other artifact on Maven Central — roughly 500,000 battle-tested libraries, all implementation away. No FFI bridges, no native compilation, no per-OS maintainers. Same Gradle dependency line as your test framework.

Where Nucleus draws the line

Two common desktop UX needs don't have a Nucleus module because the ecosystem already covers them:

• File dialogs — use FileKit.
• Spell check — use PlatformSpellCheckerKt.

The Nucleus Gradle plugin ships preloaded GraalVM reachability metadata for FileKit, so it works under Native Image with zero manual config.