Hi HN,

While working on large Java projects, I often found myself digging through heap dumps and flame graphs just to answer a simple question:

Which methods are actually slow, and how much do they cost?

Traditional profilers are powerful, but interpreting the output takes time. AI tools can read source code — but they don’t see what happens at runtime.

So I built GalataJ.

GalataJ is a lightweight Java profiler that works directly inside IntelliJ (including Community Edition) and VS Code. The workflow is simple: Profile → Compare → Act, without leaving the editor.

Profile Bytecode instrumentation attaches to a running JVM (no restart required). It collects per-method execution time, call count, and memory allocation. Metrics appear directly in the editor as CodeLens hints above your methods.

Compare You can save profiling sessions and define baselines. If a method becomes 30% slower compared to a previous run, you see it immediately — no dashboards or flame graph interpretation required.

Act Profiling results are exported as structured Markdown files inside the project (.galataj/). They are: * Plain text * Version-controllable * Tool-agnostic There’s also an “Add to Chat” action that sends the structured runtime data directly to the IDE’s built-in AI assistant. Instead of describing a performance issue manually, you can provide actual execution metrics and let the AI reason with real runtime context. The goal isn’t automation — it’s giving better inputs.

What it is not: * Not an APM * No distributed tracing * No production monitoring * No automatic deadlock detection It focuses on one thing: Helping developers understand runtime bottlenecks during development.

Would appreciate feedback from JVM and performance folks. Happy to answer questions about the instrumentation approach or overhead trade-offs.

website : www.galataj.com