Your AI Assistant Forgets Everything: Here's the Fix

You open a new chat. Again.

"It's a Next.js 15 app using App Router with tRPC and Drizzle ORM. Auth is handled by Supabase in src/lib/auth/, the database schema is in src/db/schema.ts, and all inputs are validated with Zod..."

Sound familiar? This scenario is rarer now than it used to be — but it still happens, and it's not the only problem.

The Real Problem: Discovery Is Expensive

Modern AI assistants have gotten good at exploring codebases on their own. Tools like grep, find, and Read let them crawl through your project and piece together the architecture. They'll get there eventually.

But "eventually" has a cost. Every file read burns tokens. Every search adds latency. A question like "how does auth work?" can trigger dozens of tool calls — the AI grepping, reading file after file, backtracking. You're paying for that exploration in time and money, every session.

The AI isn't broken. It's just doing archaeology every time it opens your project.

What If Your AI Already Knew?

Imagine typing: "Add password reset functionality." No preamble. No waiting while the AI reads 30 files to understand your auth setup.

Your AI responds with code that follows your patterns, uses your middleware, fits your structure — in seconds. Because it already knows.

Introducing agents-reverse-engineer

agents-reverse-engineer (are) is a CLI tool that solves this. It generates AI-friendly documentation your assistant automatically reads:

  1. Analyzes every file using AI to understand purpose, exports, and dependencies
  2. Generates .sum files — compact summaries per source file
  3. Creates AGENTS.md files — directory-level overviews aggregating child summaries
  4. Installs session hooks — automatic integration with Claude Code, Codex, Gemini CLI, and OpenCode

How It Works

npx agents-reverse-engineer@latest

One command. The installer guides you through setup. Then:

npx are init       # Initialize configuration
npx are discover   # Scan your project
npx are generate   # Generate documentation

Under the hood, are runs a two-phase pipeline: parallel file analysis producing .sum files, then post-order directory aggregation building AGENTS.md from deepest directories upward.

Session hooks inject context automatically. Updates run incrementally — are update re-analyzes only changed files using content hashing.

The Result

Your next AI session is different. Ask architectural questions and get informed answers. Request features and get code that fits your patterns.

No more explaining your codebase. No more burning tokens on rediscovery.

npx agents-reverse-engineer@latest

Fewer tokens. Faster answers. Better code.

agents-reverse-engineer — MIT licensed | GitHub