How It Works

94%

AI Detection

<50ms

Analysis Time

<1s

Context Injection

100%

Local-First

Git tracks what changed. Vreko understands who changed it. And more importantly, why.

The Journey

Timeline	What Vreko Does	You Get
Day 1	Detects AI tools, catches common risks	94% accuracy right out of the box (no tuning needed)
Day 30	Learns YOUR codebase patterns	”Cursor breaks this auth module”
Day 90	Predicts problems proactively	Issues caught before you notice

Detection

Vreko watches every save. Two systems work together:

Identifies which AI tool is active. Cursor, GitHub Copilot, Claude Code, Windsurf. We track contribution percentage too. 94% accuracy across the common ones (the ones everyone uses).

Scans for common AI-introduced risks:

Risk Type	What It Catches
Secrets	API keys, JWT tokens, database strings
Mocks	Test data in production code
Phantom deps	Imports missing from package.json

Automatic Snapshots

Vreko decides when to create snapshots based on context:

Trigger	Action
AI tool makes substantial changes	Snapshot
Critical files modified (auth, config)	Snapshot
Multiple files changed rapidly	Snapshot
High-risk patterns detected	Snapshot

Protected by default: Source code, configs, environment files, documentation.

Not protected: .git/, node_modules/, build outputs, binaries.

Sessions

Vreko groups related changes automatically:

Session: "Claude refactored auth" (14:05 - 14:23)
├── src/auth.ts      (3 snapshots)
├── src/user.ts      (2 snapshots)
├── src/index.ts     (2 snapshots) ← undefined reference!
└── AI Detection: HIGH RISK

Navigate entire sessions with one click. Not file-by-file like Git or Cmd+Z.

Protection Levels

Level	Behavior	Default Files
🟢 Watch	Silent auto-snapshots	Source code, docs
🟡 Warn	Proceed with risk boost	`package.json`, `tsconfig.json`
🔴 Block	Hard stop, no write	Lock files, `.env*`

Timeline vs Git

Aspect	Git	Vreko
Tracks	Commits you make	Every save, automatically
Granularity	Commit-level	Save-level
AI awareness	None	Full attribution
Context	Commit message	Cursor position, AI tool, risk score

Use both: Git for version control. Vreko for understanding AI-assisted development.

Privacy

100% Local-First

All analysis happens on your machine. Your code never leaves your device. Optional cloud features use metadata only-never source code.

Stored locally:

Snapshots and file states
Pattern learnings and risk scores
Session metadata

Never collected:

Source code content
File diffs
Proprietary logic

Configuration

Customize with .vrekorc:

{
  "protection": [
    { "pattern": "src/auth/**/*", "level": "block" },
    { "pattern": ".env*", "level": "block" }
  ],
  "ignore": ["*.test.ts"]
}

Recommendation: Let Vreko learn for 30 days before adding custom rules. Pattern detection improves over time.

Intelligence Format

Vreko implements the workspace.json specification for structured codebase intelligence. The daemon writes an agents.workspace.json file locally that captures fragility scores, AI modification counts, co-change patterns, and the framework manifest for your project. AI agents that support the open standard consume this file automatically — no prompting required.

Next Steps

Quick Start - Install in 2 minutes
MCP Integration - AI assistant tools
CLI Reference - Terminal commands
Troubleshooting - Common issues