Brokk, an AI Native Code Platform for Java, in Java
summary
A standalone Java Swing IDE named Brokk supervises AI coding by combining Eclipse JDT static analysis, dependency decompilation and a pure‑Java LLM runtime (Jlama) so models see type‑aware summaries instead of raw text. It uses Sonnet 46 for planning and Flash 3 for coding, runs relevant tests after every task, autocommits each step and provides undo and guided review. Benchmarks show midsize models lose ~10–15% per doubling of task size while small minis fall >50%, and Brokk measures roughly twice as fast and half the cost versus a cloud code baseline.
AI-written summary · about
chapters (12)
- 0:00Introduction and goals
- 1:55Evaluating AI coding models
- 10:11Local AI model performance
- 13:23Verification tools in coding
- 16:58Language choice for projects
- 18:01Introducing Brock platform
- 21:12Brock architecture and features
- 27:38Live demo of Brock
- 35:32Q&A session
- 41:20Code review with Brock
- 47:47Brock performance metrics
- 50:59Closing and signup
related talks
47mBrokk: an AI-Native Code Platform
01h 01mUnlocking AI's Potential: Smarter Coding in Apache NetBeans IDE
46mThe Era of AAP: Ai Augmented Programming using Modern Java
53mUnlocking AI's Potential: Smarter Coding in Apache NetBeans IDE
51mReliable AI Agents Using Domain Modeling with Koog in Java
20mAI inside the IDE
44mJava for AI
58mGenesis Keynote