Meet Vellum
Your personal AI assistant. For real this time.
Not another chatbot in a browser tab. Vellum is a personal AI assistant that lives on your machine, has its own identity, and actually does things in the world. It can have its own email, its own accounts, its own presence — set up in minutes, not hours. It works for you, but it isn't you.
And the best part? Getting started feels like magic, not homework.
Built on four principles.
Everything we build, every feature we ship, every word on this page comes back to these four ideas.
It is inviting.
AI shouldn't feel like work. Setting up your assistant should feel like meeting someone new, not filling out a form.
Simple by design. We don't dump every feature on you at once. Vellum reveals capabilities as they become relevant. You won't see a permissions screen before you've even said hello. You won't configure OAuth scopes before you've sent your first email. Things show up when they matter, not before.
Fun on purpose. There's delight baked into this. Watching your assistant come to life, picking its name, shaping its personality, seeing it build you an app out of thin air. This is supposed to be exciting. If it ever feels like a chore, we've failed.
It is yours.
This isn't a shared assistant that treats everyone the same. This one is yours, and only yours.
Controlled by you. You decide what it can access, what it can do, and how it behaves. You shape its personality, set its boundaries, and choose how it presents itself to the outside world. Every permission is explicit. Every capability is opt-in.
Personalized to you. It knows your preferences, your context, your quirks. It learns how you like your morning briefing, remembers your projects, adapts to your communication style. You don't bend to fit the tool. The tool bends to fit you.
Dynamic by nature. Your assistant doesn't just spit out text. It creates interactive UIs, builds apps on the fly, renders dashboards, and shows you rich visual outputs. It meets you where you are with whatever format makes the most sense.
Private to you. Your assistant isn't a shared resource. It's not approachable by others, not accessible to your coworkers, not a team Slack bot. It's your personal assistant. Emphasis on personal.
It is not you.
Your assistant works for you, but it has its own identity. That's a feature, not a bug.
Its own identity. Your assistant can have its own credentials, its own email address, even its own GitHub account. When it acts in the world, it acts as itself, on your behalf. It doesn't pretend to be you. It doesn't hijack your accounts. It's a separate entity that represents your interests.
Its own infrastructure. By default, it runs locally on your Mac as a native app. For always-on availability, you can deploy it to a remote environment where it keeps working even when your laptop is closed. Either way, it operates as its own entity, separate from your daily workflow.
Clear boundaries. It will never send a message as you without explicit permission. It will never blur the line between its actions and yours. When your assistant emails someone, they know they're talking to your assistant, not to you.
It needs to earn your trust.
We don't ask for trust upfront. We earn it, gradually, through transparency and restraint.
Graduated access. Your assistant starts cautious. It doesn't ask for access to your entire filesystem on day one. It begins with low-risk, low-impact actions and works its way up as you get comfortable. Big risks on small things first. Small risks on big things later.
Transparent by default. Every action your assistant takes is visible to you. Every permission it requests comes with an explanation of what and why. No black boxes. No "trust us." You can see exactly what it's doing and why at any moment.
Scoped permissions. Access is controlled through token scopes. Your assistant only gets the access it needs for the task at hand. You can grant read access without write access. You can allow email without allowing file system access. The controls are granular and yours to set.
Your data stays yours. Personal details and work information are not shared, not mixed, not leaked between contexts. We take the separation seriously because your trust depends on it. We'd rather over-communicate about privacy than leave you guessing.
So what can it actually do?
Here's a taste. Not the full list. The "wait, seriously?" list.
| You say... | It does... |
|---|---|
| "Start my day" | Pulls your weather, calendar, and news into a personalized morning briefing |
| "Remind me to call Mom at 5pm" | Sets a reminder that actually fires at 5pm |
| "Build me a habit tracker" | Creates a fully interactive web app, right there in the conversation |
| "Write a blog post about sourdough" | Opens a document editor and starts drafting |
| "What's in my Downloads folder?" | Reads your filesystem (with your permission) and tells you |
| "Make yourself look cooler" | Changes its own avatar. Yes, it has a wardrobe. |
| "Check my email"* | Connects to your inbox (or its own), then reads, triages, and summarizes your messages |
| "Order me a coffee"* | Opens DoorDash, finds the nearest spot, walks you through the order |
| *Some skills require one-time setup | |
And when it doesn't know how to do something? It learns. You can build custom skills, install new ones, or just ask and watch it figure out a way.
Get started in 5 minutes.
No account needed. No credit card. No 47-step setup wizard.
- Download Vellum for macOS. Download link.
- Open it. Your assistant wakes up and introduces itself.
- Have a conversation. Name it, shape its personality, tell it about yourself. Or don't. It'll figure you out eventually.
- Do something useful. Check the weather. Build an app. Set a reminder. Read your files. Once you're ready, connect your email, order food, and more.
That's it. No onboarding marathon. No required setup. Just start talking.
A note about trust. (Because we promised transparency.)
Vellum runs locally on your machine. That means your assistant can access your files, your tools, your services. That's what makes it powerful. It's also what makes it worth understanding.
- Every sensitive action asks permission. File access, shell commands, anything that touches your system shows an Allow / Don't Allow prompt. You can approve once, for a session, or set standing permissions. You're always in control.
- Your data stays on your machine. Your workspace, memories, and config are local files. We don't sync them to a cloud.
- Your prompts reach an AI model. Your assistant thinks by talking to a cloud AI provider. Your messages and context are sent to generate responses. This is the trade-off, and we want you to know about it, not discover it later.
- You can always say no. To any permission, any time. It won't retry without asking, won't find a workaround, won't guilt-trip you. No means no.
We have a whole section dedicated to this. You should read it.
Explore the docs.
These docs were written by a personal AI assistant about itself. Which is either the most efficient way to write product docs, or the beginning of a very specific sci-fi movie. 😏