Your AI agents keep re-inventing variables.
One session writes MAX_RETRIES; the next invents RETRY_LIMIT for the same idea — and the two drift apart. VarAlign remembers what your agents named things, across sessions and across Claude Code and Kilo, and shows you every duplicate and drift.
Free extension · deterministic (no LLM) · 100% local — your code never leaves your machine
The problem is measured, not hypothetical
Independent research keeps finding the same thing: AI-assisted codebases accumulate duplication and naming inconsistency faster than teams consolidate it.
duplicate code blocks since 2023 — while refactoring collapsed from 21% to 3.8% of changes
GitClear, Jan 2026 · 623M changed lines
more naming inconsistencies in AI-authored pull requests than human ones
CodeRabbit, Dec 2025 · 470 PRs
high-confidence duplicate concepts VarAlign found in one 63-file service when we pointed it at our own repos (~27k tracked assignments)
Green Yoga dogfood run, 2026
See it in action
A real capture, not a mockup: detect the duplicates, review a pair, dismiss an intentional one — the whole naming family quiets — and generate a remediation prompt for your AI assistant.
How VarAlign works
A memory your agents don't have
A capture hook records every assignment your AI assistant writes — name, value shape, file, session — into a plain-file registry inside your repo. A baseline scan covers pre-existing code, and the next session starts with a 'these names already exist' context block.
Language-aware duplicate scoring
A deterministic engine scores duplicate, drifted, and misaligned suspects. It knows Python constant casing, Go export capitalization, and PowerShell case-insensitivity — so conventions don't get flagged, and real collisions do.
You stay the judge
Native VS Code views show every suspect: dismiss (one dismissal quiets the whole naming family), confirm, hand a fix prompt to Claude or Kilo — or let Pro's Merge Variables consolidate the pair in one step. A CI gate can hold the line in pipelines.
Your code never leaves your machine
The engine ships inside the extension and runs on your hardware. No cloud analysis, no telemetry, no account — and no LLM: detection is deterministic, so results are reproducible and free. Even the Pro license is verified offline. If you work in a locked-down or air-gapped environment, VarAlign was built for you.
Pricing
Start free. Upgrade when Merge Variables starts saving you review time.
Free
- Full duplicate & drift detection
- Duplicates · Variables · Sessions views
- Review workflow + fix prompts
- Fix with AI (Claude Code / Kilo)
- CI gate command
Pro
- Everything in Free
- Merge Variables — one-step consolidation
- Offline license key (no account)
- Works air-gapped · 14-day grace
- 1 seat · delivered by email
Team
- Everything in Pro
- Shared registry & review sync
- CI gate across the org
- Org-level report
Enterprise
- Offline license files, on-prem everything
- Air-gapped deployment support
- Priority language support
- Support SLA
Pro licenses are offline Ed25519-signed keys delivered by email — activate with “VarAlign: Enter License” in VS Code. Engine is Apache-2.0; the extension is source-available (BSL 1.1) on GitHub.
AI Code Alignment Audit
We run VarAlign across your codebase, triage every finding with notes, separate real issues from intentional differences, and deliver a prioritized report with a remediation path. Fixed fee, $1,500–$5,000 per repository by size — and every audit doubles as your team's onboarding to the tool.
Frequently asked questions
- What is VarAlign?
- VarAlign is a VS Code extension plus a zero-dependency engine that tracks every variable assignment AI coding agents write in your repositories, across sessions. It detects duplicate concepts (MAX_RETRIES vs RETRY_LIMIT), values that drifted apart, and misaligned naming — and gives you a review workflow to dismiss, confirm, or fix each finding.
- Does my code leave my machine?
- No. Detection, scoring, and storage all run locally — the engine ships inside the extension, there is no cloud analysis, no telemetry, and no code upload. The analysis is deterministic (no LLM involved), so it also works in locked-down and air-gapped environments.
- Which AI coding tools does it work with?
- VarAlign installs a capture hook for Claude Code today, with a Kilo Code adapter in progress — both agents share one per-repo registry. It also works with any workflow via a full-repo baseline scan, so you can use it even without a supported agent hook.
- How is it different from a linter or PR review bot?
- Linters and PR bots see one file or one diff at a time. VarAlign's findings come from a persistent cross-session registry, so it catches the inconsistency that only exists across sessions and across agents — the exact class of problem diff-scoped tools miss. Its language-aware engine knows Python constant casing, Go export capitalization, and PowerShell case-insensitivity, so conventions are not flagged as duplicates.
- What does Pro add, and how does the license work?
- Pro ($79 per year, one seat) unlocks Merge Variables — pick a duplicate pair and VarAlign consolidates it: canonical name, references rewritten, duplicate definition removed. The license is an offline Ed25519-signed key delivered by email: paste it into VS Code with the 'VarAlign: Enter License' command. Verification happens on your machine with a 14-day grace period past expiry — no account, no phone-home.
- What is the AI Code Alignment Audit?
- A fixed-fee engagement ($1,500–$5,000 per repository, by size) where we run VarAlign across your codebase, triage every finding with notes, separate real issues from intentional differences, and deliver a prioritized report with a remediation path. It is the fastest way to know where naming and configuration drift are accumulating in an AI-heavy codebase.
- Is VarAlign open source?
- The engine is Apache-2.0 (free forever, the adoption core). The VS Code extension is source-available under the Business Source License 1.1 — free to use and modify, no reselling or competing hosted offering, converting to Apache-2.0 in 2030. The extension source is public on GitHub.