One app.
Choose the model that fits.
Download the app first, then a model on first launch: Lite is 2.49 GB, Med 7.30 GB, and Max 16.55 GB. Model inference runs locally without a hosted inference API key.
The macOS build is ad-hoc signed and not notarized. macOS may block opening it. Review the release and its checksum before deciding to install. A Developer ID signed and notarized release is still pending.
The Windows preview is unsigned and may trigger SmartScreen. It includes an x64 engine, but has not yet been validated on Windows hardware. Clean-machine launch and inference checks remain pending.
First launch downloads and checks the model hash. Local inference needs no Python installation or inference API key. Optional account, update, and live-search features use the network.
Windows
Download for WindowsPortable — no installer, no admin rights · not yet run on Windows hardwarePick your Laro.
Or let Veylaro pick for you.
The release artifact and each model bundle are verified independently. Veylaro reads your machine on first launch, recommends a tier, and keeps unreleased tiers unavailable.
Laro Lite release gated
The smallest Gemma 3 bundle. Start here on memory-constrained machines; speed and context headroom depend on your hardware.
Laro Med recommendedrelease gated
A larger Gemma 3 bundle for machines with more memory. Current release-attributed coding benchmarks are pending.
Laro Max release gated
The largest current Gemma 3 bundle. Requires substantial memory plus room for context, the editor, and other applications.
Install once.
Build locally.
Install the preview
Review the platform limitations and checksum, then select a model that fits your memory. Initial download time depends on your connection.
Point it at a project
Pick a file or folder. Project indexing and model inference run on-device. Optional online features are separate.
Give it real work
"Fix the flaky checkout test." "Add dark mode." Watch an agent plan, edit, run and verify — all on your silicon.
Runs on the machine
you already own.
Veylaro auto-detects your hardware and recommends conservatively: Lite below 12 GB, Med from 12 GB, and Max from 24 GB. More memory improves speed and context headroom; it does not change the local privacy boundary.
- Laro Lite — 8 GB minimum and recommended
- Laro Med — 12 GB minimum, 16 GB recommended
- Laro Max — 24 GB minimum, 32 GB recommended
- Allow additional free disk space for indexes, projects and updates
Preview status
Published artifacts have SHA-256 checksums. macOS notarization and Windows clean-machine validation are still pending.
Local inference
Model inference and project work run locally. Optional account, update and live-search features use the network only when disclosed.
Free tier inside
Every download includes the free tier — no trial clock on the model itself. Upgrade to Pro inside the app when you're ready.
Still weighing it up?
Read the historical results and current evidence limitations before downloading a model.