I build autonomous orchestration tools for AI-assisted software development. My focus: making Claude Code agents work reliably in parallel — with structure, quality gates, and measurable results.
Website: setcode.dev | Personal: gabor.setcode.dev
SET (Ship Exactly This) — Autonomous multi-agent orchestration for Claude Code.
Give it a spec, get merged features. Six pillars: SPECIFY, DECOMPOSE, EXECUTE, SUPERVISE, VERIFY, LEARN. Parallel agents in git worktrees, 7 quality gates, self-healing sentinel supervisor, deterministic output.
spec.md ──► digest ──► decompose ──► parallel agents ──► verify ──► merge ──► done
1,500+ commits · 134K LOC · 100+ E2E runs · 376 capability specs
Works on greenfield apps, brownfield codebases, and isolated units alike. SET itself is built with SET — every spec, every commit, through the same pipeline.
| Repository | What it does |
|---|---|
| set-core | Core orchestration engine — sentinel, quality gates, worktrees, memory, web dashboard |
| set-voice-agent-delivery | Voice agent — Soniox STT + Google TTS for spec-driven customer interaction |
Built-in modules: Web (Next.js, Prisma, Playwright) ships with set-core. Custom project types (fintech, healthcare, etc.) can be added as plugins.
Systems like SET can do the work of a full development team — given proper specifications and well-developed project types. This is the present, not the future.
Don't blame the model for underspecification. 90% of "AI failures" are gaps we left empty. SET exists to enforce structure (OpenSpec), verify output (quality gates), and recover from failures (sentinel + issue pipeline).
Built and used in production by ITLine Kft. · gabor@setcode.dev