Cloud Sandboxes
Isolated environments.
Real infrastructure.
Each agent gets its own sandbox — a full copy of your codebase on its own branch, running in its own cloud VM. No conflicts. No risk to your work. Close your laptop and come back tomorrow — everything’s exactly where you left it.
This is professional-grade isolation built on Git worktrees and Firecracker microVMs. Not Docker containers with shared state. Not ephemeral playgrounds that lose your work. Real infrastructure that respects the way professionals actually build software.
The problem
Without isolation, parallel agents are impossible.
You start an agent on a task. It begins editing files. But you’re editing some of those same files. Or you launch a second agent. Or the agent goes down a dead end and leaves scattered changes across your project.
These aren’t edge cases. They’re the inevitable result of running stateful processes without isolation. Sandboxes solve this the way professional infrastructure solves it: complete separation of concerns.
Agent conflicts
Agent modifies files you’re editing
Without isolation, your agent writes to the same files you have open. Your changes collide. You lose work or the agent’s work is overwritten.
With sandboxes
Each sandbox is a separate Git worktree. Your main branch stays untouched. The agent works on its own copy.
Parallel conflicts
Multiple agents step on each other
You run two agents in parallel — one on frontend, one on backend. They edit the same config file. One overwrites the other. The build breaks.
With sandboxes
Each agent gets its own sandbox with its own branch and working directory. No file conflicts, ever.
Scattered changes
Failed experiments leave a mess
The agent tries something, it doesn’t work, and now you’ve got half-finished changes scattered across your project. Rolling back is tedious.
With sandboxes
Failed experiment? Reset or delete the sandbox. Your main branch never knew it happened.
Architecture
Git worktrees.
The right way.
Every sandbox is a real Git worktree — a separate branch and working directory that shares the same repository history. This isn’t a hack or a wrapper. It’s using Git exactly as designed, with professional-grade isolation and zero overhead.
Most tools reinvent isolation badly because they don’t understand the tools they’re building on. Vera Studio uses 15 years of Git infrastructure exactly as the maintainers intended. Professional infrastructure built by people who know what professional infrastructure looks like.
my-saas-app
3 workspaces
Main branch — your active development environment. Changes here are yours.
Isolated worktree on its own branch. Agent can edit, run commands, experiment freely.
Agent finished. Review changes, merge to main, or reset the sandbox.
Enable via right-click project icon → Enable Workspaces
Cloud infrastructure
Persistent state.
Production-grade VMs.
Each sandbox runs on its own Firecracker microVM with persistent NVMe storage. Not ephemeral containers that lose your work. Not shared VMs with noisy neighbors. Dedicated compute with sub-second checkpoint and restore, powered by the same infrastructure AWS uses for Lambda.
Close your laptop, open it tomorrow — your agent is still running, your files are exactly where you left them, your environment is identical. This is what professional development infrastructure looks like when it’s not locked behind a Linux terminal.
Sub-second checkpoint & restore
Firecracker microVMs snapshot and restore in under a second. Pause a sandbox, resume it later — the process state, memory, and filesystem are exactly as you left them.
Persistent storage
Every sandbox gets persistent NVMe storage that survives between sessions. No ephemeral containers that lose your work. Your files, your dependencies, your build artifacts — all preserved.
Access from anywhere
Desktop app, mobile app, or browser — your cloud sandboxes are accessible from all three. Start work on your Mac, check progress from your phone, review from your iPad.
Agents keep running
Your agents don’t stop when you close your laptop. They keep working in the cloud. Get a push notification when they’re done. Review and approve from anywhere.
Management
Full control over every sandbox.
Sandboxes aren’t opaque black boxes. You see exactly what each one contains. Rename them to track purpose. Reset them when an experiment goes wrong. Delete them when you’re done. This is the control professional developers expect, made accessible to everyone.
- R
Rename
Give each sandbox a descriptive name so you always know what’s running where. “Auth refactor”, “Payment integration”, “Bug fix #42”.
- ↺
Reset
Discard all changes in a sandbox and start fresh from main. The agent went down a bad path? One click to undo everything.
- ×
Delete
Clean up sandboxes you no longer need. Removes the worktree and branch entirely. Keep your workspace list tidy.
Sandbox management
Multiplied power
Parallel work requires isolation.
Sandboxes are useful with one agent. They’re indispensable with five. Each parallel agent gets its own sandbox — its own branch, its own files, its own running processes. No agent can interfere with another. No merge conflicts mid-session.
This is how professional teams coordinate work: isolation first, integration later. The difference is that Vera Studio makes it work for AI agents without requiring a CS degree.
Three agents. Three sandboxes. Zero conflicts.
Professional isolation.
Zero configuration.
Persistent cloud sandboxes with full isolation. One click to provision. Close your laptop and come back — everything’s still there. This is how professional development infrastructure should work.