Getting Started with vdiff

Install vdiff, configure your API key, and run your first AI code verification analysis in under 5 minutes

Install

bash

npm install -g @4bk/vdiff
pip install graphifyy

vdiff requires Graphify (Python 3) to build the dependency graph for your codebase.

The free tier allows 15 recorded CLI events per device per day (each analysis and each completed vdiff init counts toward that cap). Upgrade to Pro for unlimited usage and offline support.

Set up

Navigate to your project and run the interactive setup wizard:

bash

cd your-project
vdiff init

This walks you through everything: LLM provider and API key, model selection, IDE integration, and knowledge graph initialization. Run this from your project root so vdiff can index your codebase.

Supported providers: Anthropic (recommended) and OpenAI.

Run your first analysis

bash

vdiff -v

That's it. vdiff detects your staged changes (or unstaged if nothing is staged) and produces a structured verification report with all sections expanded.

What you'll see

The output separates what factually changed from what might be wrong from what can't be verified:

Section	What it shows
Verdict	Merge readiness: ready, caution, or not ready
Summary	Plain-language description of the change
Impact	Blast radius: how many files depend on what you changed
Findings	Risk inferences with severity, confidence, and evidence
Missing Evidence	What can't be verified from the diff alone

Go deeper

bash

# Include full source files for better accuracy
vdiff -f

# Deep analysis with a more capable model
vdiff --thinking-level high

# Verify against a spec or PRD
vdiff -s docs/spec.md

# Branch diff (PR-style review)
vdiff --base main

← Code Health Is Necessary, Verification Is Sufficient vdiff CLI Reference→