Hosting options
Your Vellum assistant needs a computer to run on. Which computer, and who manages it, is the hosting decision. The right choice depends on what matters most to you.
There are several options, each with different tradeoffs. Some are available today, others are on the roadmap. This page explains all of them so you can make an informed choice now and know what's coming.
The three dimensions
Every hosting option is a tradeoff between three things:
Ease of use
How much setup and maintenance is required? Does Vellum handle the infrastructure, or do you?
Privacy
Where does your data live? Does it pass through Vellum's servers, or does it stay entirely on hardware you control?
Security
How isolated is the assistant from your personal files and system? What's the blast radius if something goes wrong?
No single option wins on all three. The more convenient the setup, the more you're trusting someone else with your data. The more private the setup, the more you're managing yourself.
The options
At a high level, there are three categories of hosting. Within each, there are flavors.
Local
Your assistant runs on your Mac. Your data never leaves your machine. There are three flavors of local hosting:
- Native - The assistant runs directly as a process on your Mac. This is the only option available today. It's the simplest setup, but the assistant has the most access to your system.
- Docker Coming soon - The assistant runs inside a Docker container on your Mac. Think of it as a computer inside your computer. It provides better isolation than native, but requires Docker to be installed and running.
- Apple Container Coming soon - Similar to Docker, but runs on native Apple virtual machines. No need to install Docker. You get the isolation benefits plus Apple's hardware optimizations and native security features.
All local options keep your data on your machine (great for privacy) and give the assistant direct access to your files and tools (great for power). The tradeoff: the assistant is only available when your computer is awake, and native means the assistant is running with fewer isolation boundaries.
Once Docker and Apple Container options are stable, native local may be phased out in favor of the more isolated options.
Vellum ManagedComing soon
Your assistant runs in Vellum's cloud. We provide all the infrastructure: compute, networking, storage, and managed API keys. You don't need to set up ngrok, manage a server, or even keep your computer on.
- Ease of use: The best. Minimal setup, Vellum handles everything.
- Privacy: Your data lives on and passes through Vellum's servers. If you're privacy-conscious about your data touching third-party infrastructure, this may not be for you.
- Security: Arguably one of the most secure options for the user. The assistant runs completely sandboxed in our cloud, not on any of your hardware. If something goes wrong, the blast radius is contained to our infrastructure, not yours.
You also get 24/7 availability. Your assistant is always on, even when your computer is off or asleep.
User-Hosted RemoteComing soon
Your assistant runs on cloud infrastructure that you own and manage: your GCP project, your AWS account, or even a Mac Mini running at home that you connect to remotely.
- Ease of use: The most setup required. You manage the infrastructure, credentials, and networking.
- Privacy: Your data stays on your infrastructure. Nothing passes through Vellum's servers.
- Security: You control the isolation level, but you also bear the responsibility for it.
This is the option for technical users who want full control and 24/7 availability without relying on Vellum's infrastructure.
Which should I choose?
Most people: Vellum ManagedRecommended
It's always on, there's nothing to maintain, you don't need to buy extra hardware, and we'll continue building features (like host computer use) to bridge any functional gaps between remote and local. For the vast majority of users, this is the right answer.
Privacy-conscious users: Local
If keeping your data off third-party servers is a priority, local is the way to go. Today that means native. Once available, Apple Container on your Mac (or a dedicated Mac Mini) will be the best local option, giving you both privacy and better isolation.
Technical users who want full control: User-Hosted Remote
If you want 24/7 availability, full data ownership, and you're comfortable managing cloud infrastructure, deploy to your own GCP or AWS account. It's more work to set up, but you get the best of both worlds: always-on and self-managed.
What's available today
Today, the only available option is local native. Your assistant runs directly on your Mac as a native process. Vellum Managed, Docker, Apple Container, and User-Hosted Remote options are all on the roadmap.