Engrim

This tool is a local development substrate that runs competing code implementations in parallel background Git worktrees and tests them before anything touches the working branch. When an AI coding assistant proposes a fix or a feature, the tool intercepts the task, checks out candidate branches into isolated worktrees, runs the project's real test suite and linters against each candidate, and merges only the one that passes. The developer's active code stays untouched until verification completes. The second half of the product is persistent architectural memory. Decisions, invariants and bug fixes are stored in a local SQLite database with hybrid search, then recalled across Claude Code, Cursor, the GitHub Copilot CLI and Google Antigravity, so a rule taught to one assistant is visible to the rest instead of trapped in a single tool. It installs as a Python package from PyPI and runs as a local daemon, aimed at engineers who want higher confidence in AI generated output without supervising every diff or repairing broken commits. Four tiers are published. The core edition is free forever under the MIT license and provides the lightweight memory layer on its own. The individual professional edition is $15 per month per engineer and includes a 14 day evaluation trial. The team edition is $20 per license per month with a minimum of five licenses, which the site states as a $100 per month base, and adds team synchronization and shared rules. The top tier is a custom contract with a commercial license and a custom enterprise end user license agreement. The site labels the current enterprise build as version 0.1.0, so the parallel worktree engine is early and should be evaluated against a real repository before it is relied on.