Code Review is the New Bottleneck.
Know if your change is safe to merge in seconds. Deterministic checks + AI reasoning. No guesses.
Try it before your next commit
See how it works →+ export const processData = async (data: any) => {
+ // AI generated code
+ const result = await db.query('SELECT * FROM users');
+ data.forEach(item => {
+ if(item.id) {
+ processItem(item);
+ }
+ });
+ return true;
+ }
+ export const processData = async (data: any) => {
+ // AI generated code
+ const result = await db.query('SELECT * FROM users');
+ data.forEach(item => {
+ if(item.id) {
+ processItem(item);
+ }
+ });
+ return true;
+ }
~/projects $ ▊
→ Know if it's safe
→ See what breaks
→ Get clear next steps
The Trust Gap
+60%
higher defect risk when AI works on unhealthy code
Code for Machines, Not Just Humans, 2026
41%
more bugs from AI adoption with no throughput gain
Large-scale industry study, 2025
-19%
actual time lost despite devs believing AI saved them 20%
Controlled developer study, 2025
Research shows AI coding tools increase defects without verification.
“35% of developers access AI tools through personal accounts. We had no visibility into what was being generated or reviewed.”
VP of Engineering
“Our auditor asked how we verify AI-generated code. We didn't have an answer.”
Head of Security
“AI-generated code passes tests but silently removes error handling, widens auth boundaries, and drops null checks.”
Staff Engineer
How it works
From install to insight in under 5 minutes
1
Init
Configure your LLM provider, API key, and let vdiff understand your codebase. One command.
vdiff init
2
Analyze
Point vdiff at your staged changes. It parses the diff, reasons about risk, and generates useful data points for you about your code.
vdiff -v
3
Review
Get a structured report with facts, not opinions. See what changed, what's risky, and what's missing before you merge.
# vdiff analysis report
The Signals You Get
One command. A structured report that tells you exactly where to look before you merge.
VERDICT
Merge or don't
An immediate decision: Ready, Caution, or Not Ready. Based on a computed risk score across your entire change set. No guessing.
IMPACT
Blast radius
See exactly how many downstream files depend on what you changed. Know what you're touching before you merge.
FINDINGS
Prioritized risks
Every finding has a severity, a confidence level, evidence, and the exact code that triggered it.
SPEC
Spec alignment
Point vdiff at the spec and see which requirements were actually fulfilled. Get insights on addressed, partially addressed or missing requirements with real evidence.
FOCUS
Where to look
A prioritized list of the exact files and lines that need your attention. Skip the boilerplate. Focus your cognitive budget on actual risk.
DELTA
Review memory
vdiff remembers previous findings. See what's resolved, what's still open, and what's new since your last review. Optimized for fast iterations and rapid feedback loops.
Not another AI reviewer
Code stays on your machines
No vendor servers, no remote indexing, no code leaving your infrastructure. Only the diff reaches the LLM - using your own API keys.
Evidence, not opinions
Every finding comes with severity, confidence level, and evidence from the actual diff. Facts are separated from inferences. Uncertainty is explicit.
Deterministic + AI
Tree-sitter AST analysis catches structural changes mechanically - immune to prompt injection. The LLM layer adds reasoning on top, clearly labeled as inference.
Audit-ready output
Structured JSON reports, CI exit codes, finding history, and review memory. Every analysis is stored and replayable. Built for teams that need to answer 'how do you verify AI code?'
From your terminal to your CI pipeline
One tool, every stage of review.
Before you commit
Catch what tests miss.
Before you merge
Block risky changes.
Before it's too late
Scale review with confidence.
Pricing
Personal
Free
For individuals and open source projects.
✓
All features unlocked
✓
Bring your own LLM keys
✓
No account required
Pro
Coming soon
For teams and organizations.
✓
Team dashboards
✓
Shared review history
✓
CI enforcement
✓
Priority support
Stop reviewing everything. Review what matters.
Give vdiff a try. Focus on what matters.
Works with your stack
Just run vdiff.
TypeScript
.ts .tsx
JavaScript
.js .jsx
Python
.py
Go
.go
Java
.java
Rust
.rs
Ruby
.rb
C
.c
C++
.cpp
C#
.cs
Kotlin
.kt
Swift
.swift