Turbo Rig

private beta

The lab's integrated agentic coding rig

A governed rig for AI coding agents: 8 thin scripts, zero daemons, a cross-family code-review gate where a different model family reviews every diff, git-versioned agent memory, and human-only merges. Currently in private beta.

Repository private while in beta; runs in production inside the lab daily.

Language
Bash, Python
License
MIT
Version
v5
Released
2026-10-07

How it works

Agents build in isolated worktrees; every PR passes a fail-closed review gate run by a different model family; humans are the only mergers. The gate applies a published finding: in a controlled study of 116 tasks (Xiang et al., arXiv:2607.21656), cross-model review lifted pass rates from 71.6% to 89.7%, outperforming same-family self-review.

Links