Advanced JVM obfuscation,
zero-overhead native execution.

NekoObfuscator is an open-source Java bytecode and native obfuscation toolkit that rivals commercial tools like ZKM — with deep HotSpot integration that bypasses JNI entirely, delivering native performance indistinguishable from raw speeds.

Every layer counts.

Eight independent transform stages that stack into a hardened protection chain. Each operates at the bytecode level — no source modifications required.

01

Renamer

ZKM-style class, field, and method renaming with mapping output and automatic reflection rewriting.

02

Key Dispatch

Hidden long-key injection into method signatures with automatic call-site rewriting.

03

Control-Flow Flattening

Keyed island dispatchers over verifier-safe blocks with per-edge evolving keys.

04

Runtime Variable Obfuscation

CFF-live keyed shadow storage for primitives — no ThreadLocal warehouse needed.

05

Constant Obfuscation

Numeric push, LDC, IINC, and ConstantValue rewriting bound to live CFF state.

06

String Encryption

AES/DES + XOR stream mixing with class-local cipher caches. No helper classes.

07

InvokeDynamic

CFF-state keyed indy indirection that eliminates static reference graphs from the constant pool.

08

Native Translation

Bytecode-to-C translation via Zig with direct HotSpot method-entry patching.

Not JNI. Not JVMTI.
HotSpot.

Every other native obfuscator pays the JNI tax. Neko patches HotSpot directly, achieving native execution speeds within measurement noise — and sometimes faster than — raw, un-obfuscated code.

01

HotSpot entry patching

During JNI_OnLoad, Neko patches vtable/itable entries in HotSpot's Method structures, redirecting execution to compiled native code without JNI overhead.

02

Zero JNI boundaries

Traditional native obfuscators cross JNI on every call — marshalling, handle management, JNIEnv lookups. Neko bypasses all of it.

03

Indistinguishable performance

Native execution within measurement noise of raw JVM speeds — and faster in compute-heavy workloads. No soft fallback, strict barriers or abort.

04

JDK 21+ multi-platform

Fresh runtime validation for each target JDK, platform, and GC mode. Compiled via Zig for Linux x64, with cross-platform support expanding.

Traditional JNI Path
Java → JNI stub → Marshall args → Native exec → Unbox result → Java
vs
Neko HotSpot Path
Java → Patched vtable → Direct native call

Built by autonomous agents,
not boilerplate generators.

Every line of NekoObfuscator — from the bytecode IR to the HotSpot patching layer — was synthesized by a multi-agent harness. One agent architects the system and generates the heavy-lifting code; another verifies safety, models JVM internals, and orchestrates native compilation. No human wrote a single transform.

This isn't AI-assisted coding with autocomplete. It's autonomous software engineering — agents that reason about JVM specs, design IR lowering passes, and write cross-compilation pipelines. The result rivals commercial obfuscators in structural complexity, built entirely by LLMs collaborating through a structured harness.

100% AI-synthesized
codebase
Architect & Generator
Verifier & Orchestrator

Near-zero overhead.
Sometimes faster than raw.

By bypassing JNI entirely and compiling to native via Zig, Neko eliminates the marshalling tax that cripples traditional native obfuscators. In compute-heavy workloads, the native code can even outperform the JVM JIT.

Calc benchmark Static-only, arithmetic-intensive loop
9–12ms Raw JVM
2–3ms Neko Native
Matrix multiplication 4096 × 4096, double-precision
3–4ms Raw JVM
9–10ms Neko Native
JNI-based obfuscator (reference) Equivalent workload, standard JNI boundaries
3–4ms Raw JVM
150ms+ JNI Obfuscator

* Benchmarks run on Linux x64, JDK 21+, G1 GC. Neko native compiled via Zig with -march=x86_64_v3. JNI reference values are typical for standard JNI-bound native obfuscators — the overhead comes from argument marshalling, JNIEnv lookups, and object handle management on every call boundary.