A body for your agent
A root box boots in seconds. Your Claude, Cursor or Codex connects and works like it owns the machine — except it isn't your machine. Prototype survived? Crystallize it into a permanent VM with a domain.
You bring your own agent. We don't resell tokens.
A laptop is a bad body for an agent
The agent doesn't get tired, isn't afraid of breaking the system and will happily work all night. The constraint isn't the agent — it's the machine it runs on.
The third agent kills your laptop
One agent is fine. Three in parallel means fans at full tilt, builds queued and an unresponsive IDE. Parallelism hits the hardware you carry around.
Agents leave a mess behind
Four Node versions, global packages, 12 GB of images and a systemd unit you'll discover next month. Your own machine is too expensive to risk.
Nothing to show
The prototype lives on localhost. To hand a client a link you need an address, TLS, a database and somewhere that survives closing the lid.
How it works
Four steps. The first takes seconds, the last turns an experiment into production.
dada box upA body in seconds
A warm root box with node, python, go, docker, git and compilers already inside. The agent doesn't spend its first minutes on apt install.
agent connectsYour agent, our body
Claude Code, Cursor or Codex stay with you; tools execute in the box. We don't host the model and never see your subscription.
dada box attach db / s3Resources mid-flight
Need a database or a bucket? They attach to the running box, with no migration and no config rewrite. Managed, with backups.
dada box crystallizeCrystallize into a VM
The prototype survived — the same object becomes a permanent VM with a domain and TLS. Not a rebuild, not a migration: the same environment, continued.
One object, from thought to production
Today a sandbox and production are different products in different formats, and moving between them is a rewrite. Here it's one environment whose lifecycle stage is a property, not a species.
# play demo →
Crystallization
The real difference from a sandbox: the experiment is neither thrown away nor rebuilt. It moves across whole, and grows up. One identity for the whole journey.
This is a mechanical move of an object, not a rebuild from a description. No model participates in this path — so there is nothing to guess wrong.
What carries over
- the box filesystem, exactly as it is
- volumes and data byte for byte, with no dump-and-restore (a database inside the box needs a short pause for the final sync)
- environment variables and secrets
- attached database and bucket, same connection strings
- the public address — from a temporary one to your domain
- the same ports, and the same processes relaunched once — same commands, same working dirs
Why not just a VPS
Honest objections, honest answers.
— I'll grab a $5 VPS and be done.
For a single agent, reasonable. For three in parallel you need three, and they idle 90% of the time. A box bills active minutes: idle costs nothing.
— I already have a server.
Then you have one body that agents slowly pollute and that you're afraid to break. The point of a box is disposability: destroy it and boot a new one in seconds.
— The agent can set up the environment itself.
It can — spending the first minutes of every session and your tokens doing it. A warm image removes that tax.
— Where does production live then?
Same place. Crystallization turns the box into a permanent VM with a domain, with no migration. That's the whole point: one environment from thought to production.
Pricing hypothesis
The numbers below are a hypothesis this preview is testing. The form asks what you think it should cost.
Box
Idle isn't billed. A forgotten box goes to sleep instead of burning your budget.
Database and S3
Managed Postgres and object storage with backups — the same ones as in the console.
Crystallized VM
A permanent body, domain, TLS. Comparable to a VPS plus managed services.
Preview access is free and granted by hand. Billing won't switch on until we know boxes are actually being used.
Questions
No, and we don't plan to. You bring your own agent and your own subscription. We provide the body it works in. That way there's no token markup and no access to your subscription.
What works today and what doesn't
This is a private preview, not a finished service. We'd rather you didn't discover the product's edges on your own project.
Works today
- managed Postgres, S3, domains and TLS — our core platform, in production
- provisioning and configuring permanent VMs — running customers rely on it
- MCP access from an agent to platform resources
- preview boxes — provisioned by hand, by an operator, after your request
Not yet
- self-service: a box doesn't boot on a button press without us
- one-step crystallization — today it's several operations with us involved
- per-minute billing
- start-time guarantees: we'll advertise seconds once it's always true
- zero-downtime promotion: there is downtime, it is short — tens of seconds, longer on heavy data. We quote the number rather than claiming "no restart"
- moving processes without a restart: we're measuring demand, but today we consider it honestly out of reach
Request preview access
We grant access by hand, in small batches. Tell us what you plan to run so we can prepare a box for your scenario rather than a generic one.