RACT (Root Agentic Coding Tool)
An Agentic Coding Tool built around a small management LM, by Dr. Lucas Root, Ph.D.
RACT keeps the human in the loop while letting a lightweight core manager route work to the right LLM provider. Every operation is anchored to the assumption that justifies it through the Rooted[T] signature quirk.
Want the thinking behind RACT? Dr. Root shares the deeper philosophy — and a free chapter — in the AI Agent Playbook.
Install
pip install rootact
Or scaffold a project in one command:
rootact init --template python-package --provider local
What makes RACT different
- Model-agnostic — use local models, OpenAI, Anthropic, Z.ai, Moonshot, OpenRouter, or any OpenAI-compatible endpoint.
- Root-Knot-anchored — every file carries the author’s identity markers; unsigned work cannot compound.
- Self-recursing loop — the Progress Oracle plans milestones, verifies them, and stops only when the work is done or a regression is detected.
- Operator Handshake — high-risk actions queue for review instead of pausing the loop.
- Anti-rot guardrails — duplication guard, refactor tax ledger, error-mask detector, novelty budget, Chesterton’s Fence, Dead Code Auction, and Legacy Whisperer.
Quick links
From the author
RACT is the public, standalone expression of ideas I’ve been developing around assumption-driven programming and model-agnostic agentic tooling. If you want the longer-form thinking behind it — including how to design agentic systems that stay accountable, auditable, and genuinely useful — I share that in my AI Agent Playbook. The first chapter is free.
License
RACT is licensed under the PolyForm Noncommercial License 1.0.0 — free for personal use, research, education, and noncommercial organizations.
Commercial use requires a separate agreement. See COMMERCIAL.md for details, or email info@lucasroot.com.