--- title: "Checkpoint & Restore — Sprites" description: "Statefulness, with an undo button. Checkpoints on Sprites." --- # Checkpoint & Restore Statefulness, with an undo button Sprites are safe, sandbox environments with checkpoint and restore. Restore your entire environment to a working state in seconds. $30 trial credits create, like, 500 Sprites for free [Play with Sprites](https://fly.io/app/sign-up) [Docs](https://docs.sprites.dev/) ### Footgun-proof vibe coding AI agents can write goofy code. Sprites take the stress out of vibe coding by letting you create and restore from checkpoints. ### Version control for the whole VM Creating a checkpoint on a Sprite is like a git commit for your whole system. Database changes, packages, policies: it all gets tracked in a checkpoint. ### Fast and secure backups Checkpoints take about 300ms to create and your Sprite environment won't even notice. Backed up to object storage for durability and peace of mind. Quick Start \# Install the Sprites CLI curl https://sprites.dev/install.sh | bash \# Create a Sprite sprite create my-project \# Do some work, then checkpoint sprite checkpoint -s my-project -c "looking good so far" \# Oops. Restore to the last checkpoint sprite restore -s my-project Disposable Compute Sprites get VMs when they need them. Fire one up, do some work, forget about it for 6 months. We make sure it gets compute when you come back. Firecracker VMs Sprites execute code in Firecracker VMs. Even we have a hard time seeing what they're doing. Isolated Networking VMs run on isolated networks. Nothing can connect to your Sprite directly. Dynamic Resources Sprite VMs have access to big chunks of CPU and Memory. Your blobs of code just use what they need, up to 8 CPUs and 16GB of RAM for any given run. Tiered storage You read and write to fast, directly attached NVMe storage. Your data then gets written to durable, external object storage. Same disk for every run When Sprites become inactive, we remove their compute (and stop charging you!). When we assign new compute to Sprites, they come up with the exact same data on the filesystem. Max Compatibility Since it's just ext4, you won't run into weird edge cases like you might with NFS or FUSE mounts. You can happily use shared memory files, for example, so you can run SQLite in all its modes. Only pay for actual data You don't pay for allocated filesystem space, just the blocks you write. And it's all TRIM friendly, so your bill goes down when you delete things. Live Checkpoints Checkpoints are transactional metadata. They take about 300ms and your Sprite environment won't even notice. Copy-On-Write We only copy blocks that have changed. This keeps your bills down, incremental checkpoints don't use much space. Full Environment If you've installed packages, services, or policies in your Sprite, these are checkpointed with your code and data. Quick restores Restores put everything in place then restart your Sprite. You'll be reconnected in well under a second. Manage them within the environment Point your LLM at /.sprite/llm.txt and it'll happily take care of checkpoints for you. URLs for External Access Sprites have independent URLs. Automatically Activate Sprite Requests to your Sprite URL assign new compute to your Sprite, if inactive. Your data remains, and even a "cold start" request will typically take less than 1s. Just listen on 8080 Requests to your sprite URLs are proxied to whatever you have listening on port 8080 within the Sprite. Public Access, if you want Just run `sprite url -s update --auth public` and we open your Sprite URL to the internet. Serve up an entire app, show your friends, and accept webhooks. If it speaks HTTP it can speak to your Sprite. Fine-grained network egress Layer 3 network policies restrict outbound connections. Simple rules Configure sprites with a list of domain names to allow, include our default list of LLM friendly destinations, or let them access everything. Live updates via REST API Network policies are updated via an authenticated REST API. Updates are enforced immediately. Transparent to Applications Since these are enforced at L3, apps can use the full range of TCP and UDP protocols available to them. ## Billing Hourly Per Second You pay only for Storage and compute resources while a Sprite is active. Usage is aggregated across your organization. ### CPU Time $0.0385 /vCPU/hr Minimum 0.0625 CPU while running. ### Memory Time $0.021875 /GB RAM/hr Minimum 256MB RAM while running. ### Storage $0.50 /GB/month Prorated to the hour. 1,000 sprites using 10MB data = 10GB total. All compute resources are billed per second. Storage is billed hourly but aggregated monthly.