Design partners
A computer for every agent,
in your own cloud.
Firecracker microVMs for agents and untrusted code. Boot in about a second, pause with memory, resume in under a second, and keep the disk for next time. One API, on compute you own, in your VPC.
- cold boot to a running sandbox
- ~1 s
- resume from pause, memory and all
- <1 s
- per sandbox, never shared
- 1 kernel
Trusted by engineering & platform teams at scale
Agents are only as good as their computers.
Your code leaves your cloud
Hosted sandboxes run your repos, secrets, and data on someone else's infrastructure.
Every session starts cold
Clone. Install. Build. Again. The agent waits, and you pay for it.
A shared kernel is a thin wall
AI-written code in a container shares a kernel with everything else on the host.
With Ravion
One API.
Instant computers.
Create, exec, pause, resume. From TypeScript or plain HTTP.
Step 1
Create a sandbox in one call
Pick CPU, memory, and a disk. Name the disk and the next sandbox starts with everything the last one installed.
Replaces E2B, Daytona, Modal
Step 2
Everything is an API
Exec commands, write files, pause, resume. Pause freezes memory and disk. Resume picks up mid-thought, in under a second.
Replaces custom VM orchestration
Step 3
The same sandboxes run your CI
Pipelines and GitHub Actions runners use the same microVMs and sticky disks. One fleet, in your account.
Replaces separate CI fleet
Three calls to a computer.
Create, exec, pause. The same API from TypeScript, Python, or curl. Proposed, and open for feedback.
Untrusted code. Your perimeter.
Agents run where your data already lives. Ravion only hands out the machines.
- UI
- CLI
- API
- Git
- Schedules sandboxes
- API, SDK, and keys
- Logs and usage
- sbx_7Hq2
- sbx_K91d
- sbx_p4Wm
A network namespace per sandbox
- Postgres
- Internal APIs
A hop away
- Snapshots and disks in S3
- Scoped credentials per sandbox
Fast, isolated, and yours
One-second boots.
Firecracker microVMs on hosts in your account.
Sub-second resume.
Pause with memory. Resume exactly where it stopped.
Instant sticky disks.
Copy-on-write disks stream blocks as you read them, so a warm disk is ready before you need it.
Hardware isolation.
A VM per sandbox, its own network namespace, scoped credentials.
In your VPC.
Reach private databases and APIs without opening a single port to the internet.
Spot pricing.
Run on Spot capacity at AWS Spot rates.
Coming from another sandbox?
The Sandbox API is in early access with design partners. Here is how your code maps to it.
Teams come to Ravion from E2B when agents need to reach private data and the sandboxes have to run inside their own AWS account.
Before: import {Sandbox} from "e2b"
After: import {Ravion} from "@ravion/sdk"
After:
After: const ravion = new Ravion()
Before: const sandbox = await Sandbox.create()
Before: await sandbox.commands.run("pnpm test")
After: const sandbox = await ravion.sandboxes.create({
After: environment: "agents-prod",
After: })
After: await sandbox.exec("pnpm test")Move in three steps
Become a design partner and connect your AWS account.
Swap the E2B client for the Ravion SDK.
Name a disk so each sandbox starts with everything the last one installed.
Sandbox questions
Give your agents
somewhere safe to work.
We're building Ravion Sandboxes with a small group of design partners.
Firecracker microVMs / Sticky disks / Runs in your AWS account