Tailor a CV to every application without losing your history.
A client-side, privacy-first CV builder for the web. No backend, no account, no tracking. Everything you type stays in your browser.
Features · Getting started · Architecture · Contributing
Most CV tools give you one CV and let you edit it into the ground. Every time you tailor it for a role, you lose something you might have wanted next time.
CVForge splits that into two jobs.
The Vault is your master record. Every role, project, bullet point, skill, publication and qualification you've ever had, whether or not it belongs on any particular CV. It's never the thing you send.
The Studio is where you build one CV for one application by picking from the Vault. Toggle sections, choose which bullets make the cut, reorder them, and watch a real PDF update as you go.
Delete a bullet from a CV and it's still in your Vault. Build twenty CVs off one history without copy-pasting between documents.
Work history with per-role bullets, projects, skills grouped into categories, education, publications, references and hobbies. Bullets can be linked to skills, so picking a skill for a CV can pull in the evidence behind it.
Pick sections and individual bullets, reorder them, and preview the result as
a real PDF (that's the screenshot at the top). The preview is the export:
Studio rasterises the same bytes that Export PDF writes, so what you see
can't drift from what you send.
Every CV you build lands in a grid with its first page rendered, so you can tell twenty near-identical CVs apart at a glance.
 |
 |
| The Vault, with an entry open | The CVs you've built so far |
Exports are vector PDFs with selectable, copyable text, not a picture of a CV. Two templates ship today. Both are single-column on purpose, since multi-column layouts are what most often confuse a parser.
| Template | Shape |
|---|---|
| Compact | Clean, single-column, sans-serif. Built to pass through ATS parsers without friction. |
| Traditional | Centred and whitespace-led, with a two-row entry header and a justified summary. |
CV conventions also aren't universal, so you pick a region and CVForge adjusts page size, expected length, date format, what the document is even called, and the guidance it shows you. Eight regions ship: UK & Ireland, US & Canada, Australia & New Zealand, DACH, Nordics, Europe (international), Mexico/Colombia/Chile, and Brazil & Southern Cone.
 |
 |
| Templates, previewed against your own CV | Region conventions |
Upload any existing CV, not only one CVForge made, and see what a resume parser sees. X-Ray flags the format problems that make extractors misread a document. The main offender is a multi-column layout, where two columns sharing a baseline get merged into one nonsense line. Reading order traces the path a position-sorting extractor actually takes through the page.
- Tailor with AI (beta). Paste a job ad and have a model select and rewrite bullets for it. Bring your own API key; Anthropic and Google Gemini are supported. Stays off until you add one.
- Google Drive sync. Mirrors your Vault and CVs to your Drive's hidden app folder, so a second browser picks up where you left off. Stays off until you sign in.
- JSON backup. Export everything to a file and import it back.
By default CVForge makes no network calls with your data at all. There's no server component to this project. Your Vault and CVs live in your browser's IndexedDB, and PDF generation happens on-device.
That cuts both ways, so it's worth being blunt: your data is only as durable as your browser's site storage. Clearing site data, or working in a private window, will lose it. Turn on Drive sync or take a JSON backup if that matters to you.
Two optional features change the picture, and only if you turn them on.
| Feature | What leaves your browser | Where it goes |
|---|---|---|
| Tailor with AI | The job ad, plus your CV's career content: experience, projects, skills, education | Straight to Anthropic or Google, using your API key |
| Google Drive sync | Your Vault and CVs, as the same JSON the backup export produces | Your own Google Drive, hidden app folder |
Neither routes through a CVForge server, because there still isn't one.
For AI tailoring specifically, your identifying details (name, email, phone, location and profile links) are stripped before the request is built and are never part of it. The feature shows you exactly what's about to be sent, before it sends it.
You'll need Flutter 3.47.0, which is what CI pins, though newer stable releases will usually work.
git clone https://github.com/IanHeinrich/CVForge.git
cd CVForge
flutter pub get
flutter run -d chromeThat's the whole setup. There are no environment variables, no services to stand up and no API keys. The app runs fully featured without them, minus the two optional integrations above.
To build a release:
flutter build web --releaseDeploying under a sub-path, as GitHub Pages does, needs the base href:
flutter build web --release --base-href /CVForge/Flutter Web, Stacked MVVM, with freezed for immutable models.
Four things are worth knowing before you read the code.
It's organised by feature, not by layer. Views, dialogs and
feature-specific widgets live together under lib/features/<feature>/. Only
genuinely shared things stay global: services, common widgets, routing and DI.
Stacked's routing and locator are centralised by design, so splitting those up
isn't on the table.
Models never import Flutter. lib/models/ is pure Dart.
render/cv_composer.dart is the only place Vault data and Draft selections get
joined into the ResolvedCv a template renders.
There's one renderer, not two. A template owns exactly one pdf-package
renderer. Studio's live preview rasterises that same PDF instead of keeping a
parallel Flutter widget tree alive, so preview and export can't disagree on
content or on pixels.
Pagination is already solved, and it's a trap. Splitting a CV across pages
is harder than it looks in package:pdf. templates/design/section_pagination_pdf.dart
is the one place it's done correctly, and new templates have to route through
it. CONTRIBUTING.md explains why before you write one.
lib/
├── app/ # Stacked wiring: routes, locator (generated)
├── features/ # Vertical slices
│ ├── analyzer/ # ATS Check
│ ├── settings/
│ ├── studio/ # Building one CV
│ └── vault/ # The master record
├── models/ # Pure Dart, freezed. No Flutter, no pdf.
│ ├── draft/ # What a CV selects from the Vault
│ ├── render/ # Vault + Draft -> ResolvedCv
│ └── vault/ # The master record's shape
├── services/ # Always global, even single-consumer ones
├── templates/ # One pdf renderer per template, plus design tokens
└── ui/ # Shared chrome: theme, tokens, common widgets
CONTRIBUTING.md has the full conventions and the reasoning behind them. The short version:
stacked generate # freezed, json_serializable, locator, router
dart format . # always, immediately after generating
flutter analyze
flutter test --exclude-tags=goldenNote
Golden tests are baselined on Linux, so they show small pixel diffs on macOS
or Windows even when nothing has really changed. --exclude-tags=golden is
the signal that's actually green locally. CI verifies them properly on every
PR.
Pushing to main runs deploy.yml. It looks
for a version bump in pubspec.yaml, builds with --base-href /CVForge/,
publishes to GitHub Pages and tags the release. No bump means no deploy.
404.html handles deep links. GitHub Pages has no rewrite rules, so it encodes
the requested path into a query string and index.html restores it before the
router reads the URL.
Issues and pull requests are welcome. CONTRIBUTING.md covers setup, the conventions this codebase enforces, and what CI checks.
One rule is worth stating up front: scaffold with the Stacked CLI rather than by hand. Hand-written views, services and widgets miss the route and DI wiring the CLI maintains, and drift from the generated conventions.
PolyForm Noncommercial License 1.0.0, Copyright 2026 Ian Heinrich.
Source-available, not open source: you may use, modify and redistribute CVForge for any noncommercial purpose — personal use, hobby projects, study and research, and use by charities, schools, public research bodies and government institutions. Commercial use needs a separate licence; open an issue to ask.
Built with Stacked, freezed, pdf and printing, Hive CE and Remix Icon. Deep-link handling on GitHub Pages follows spa-github-pages.