Model-agnostic, local-first agentic coding with signed provenance and assumption-driven planning.
Install
Or scaffold a project in one command:
ract 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.
- Provenance-anchored artifacts — every generated file carries a signed rootknot that binds it to the plan step, assumption, and generator that produced it.
- Assumption-driven programming — every plan step declares the assumptions that justify it; violated assumptions propagate and trigger targeted re-planning.
- Milestone-halting recursion — the loop continues only while measurable progress is being made; it stops on completion, regression, budget exhaustion, or provenance violation.
- Operator Handshake — high-risk actions queue for review instead of pausing the loop.
Quick links
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.