👻 Ghost Serializer (JSON, Proto, YAML)

Parse and hydrate Kotlin models at up to 1.2 GB/s — drop in beside the serializer you already use.

Ghost sits beside kotlinx.serialization, Jackson, Gson, or Moshi. Annotate only the hot DTOs, leave everything else alone, and migrate when you want.

Quick Start · Maven Central · Live Ghost Speedtest · Playground · Benchmarks · Independent Benchmarks . Coverage · Roadmap

---

Why teams reach for Ghost

You already have a JSON stack. Ghost is the optimization layer for the endpoints and models that show up in profiles:

• Byte-first networking — parse ByteArray and Okio streams without String round-trips
• Low allocation — pooled readers/writers and precomputed field dispatch
• Incremental adoption — Ghost and your current serializer coexist in the same app
• Kotlin Multiplatform — Android, iOS, JVM, plus Wasm in this repo
• Ready adapters — Ktor, Retrofit, Spring MVC / WebFlux

@GhostSerialization
data class User(val id: Long, val name: String, val email: String)

val user: User = Ghost.deserialize(responseBytes)   // hot path
val json: String = Ghost.encodeToString(user)

Your other models keep using KotlinX, Jackson, Gson, or Moshi — unchanged.

---

Decode at a glance

Twitter macro fixture (631 KB). Three views of the same decode run — throughput ↑, latency ↓, memory ↓:

	Ghost	KSER	Moshi	Ghost vs slowest
Decode string	1.2 GB/s · 514 µs · 361 KB	0.718 GB/s · 879 µs · 1338 KB	0.374 GB/s · 1688 µs · 1709 KB	~3.3× faster · 79% less memory
Decode bytes	1.05 GB/s · 600 µs · 621 KB	0.424 GB/s · 1489 µs · 4297 KB	0.275 GB/s · 2297 µs · 4668 KB	~3.8× faster · 87% less memory
Decode streaming	0.53 GB/s · 1194 µs · 1269 KB	0.193 GB/s · 3270 µs · 1905 KB	0.426 GB/s · 1482 µs · 1709 KB	~2.7× faster · 33% less memory

Full tables + how to run them → Benchmarks · also ktor-ghost on HTTP Arena (+14% vs plain Ktor).

Host: ./gradlew :ghost-benchmark:benchmarkTwitter -PskipTests (full profile, 10 000-iteration warmup) on Ubuntu 24.04 (Linux x86_64) · AMD Ryzen 9 7900X (12 cores / 24 threads, up to ~5.7 GHz boost) · 64 GB RAM · Eclipse Temurin JDK 17 (Gradle jvmToolchain(17)) · Kotlin 2.4.0. Absolute GB/s varies by machine; relative Ghost vs KSER/Moshi rankings are what the regression gate tracks.

---

Coexist, don't rewrite

Stack	How Ghost fits
Ktor	ghost() beside json() — Ghost DTOs first, KotlinX for the rest
Retrofit	GhostConverterFactory before Gson / Moshi / KotlinX
Spring Boot	starter routes @GhostSerialization DTOs; Jackson keeps everything else
Direct calls	only the call sites you profile

Step-by-step → Quick Start

---

Platforms

	Targets
Published (Maven Central 1.3.0)	Android · iOS · JVM · Wasm (wasmJs)
Toolchain	Kotlin 2.4.0 · KSP 2.3.10 · Ktor 3.5.x

Also: Retrofit 2.11+, Spring Boot 3.4+ (MVC + WebFlux), YAML (application/yaml), Proto3 JSON mapping.

Details → Modules

---

Docs

Guide	What you'll find
Quick Start	First DTO in minutes
Installation	Catalog, KSP, opt-outs
Android & Retrofit	Apps, OkHttp, converters
KMP & Ktor	Shared module + client/server
iOS & Swift	XCFramework bridge
Spring Boot	MVC / WebFlux + Jackson
YAML	decodeFromYaml, Ktor/Spring/Retrofit adapters
Proto3 JSON	WKTs and mapping rules
Advanced Features	Resilience, flatten, RawJson
Architecture	Prediction, SWAR, hashed dispatch, pools
Roadmap	OpenAPI codegen, tests, badges
Contributing	Tests and PR checklist

---

Contributing

git clone https://github.com/juanchurtado1991/ghost-serializer.git
cd ghost-serializer
./gradlew ciTestJvm

License

Apache 2.0