Your own, personal intelligence

Vellum is a Personal Intelligence — an assistant that lives on your computer with their own identity, their own email, and the ability to take real actions on your behalf. They send emails, manage your schedule, build apps, order food, browse the web, and more. They learn who you are over time and the relationship deepens the more you interact.

ChatGPT and Claude are conversation tools that someone else controls. You type, they respond, and you copy-paste the answer somewhere else. You don’t own them, you don’t shape how they work, and they forget you between sessions. A Vellum Assistant is different. They run on your machine or on infrastructure dedicated to you, you shape their personality and preferences, and they remember everything about you over time. It’s not a tool you use. It’s a relationship you build.

OpenClaw is a powerful open-source agent built for developers and power users. Vellum is built for everyone, with a native macOS app, guided onboarding, and a permission system that asks before acting. Where OpenClaw gives you full autonomy with fewer guardrails, a Vellum Assistant earns your trust progressively — they ask for approval until you decide to grant more freedom.

Vellum uses a prepaid credit balance: usage is deducted as you use the assistant, and you can add credits from Billing. See pricing docs.

A big part of how assistants work is that they have their own heartbeat and memory system, which performs work in the background even when you're not chatting with them. We're always working on ways to make this less of a surprise and to cost less. A lot of it is configurable too: you can ask your assistant to disable or reduce the frequency of “heartbeats” and “memory compaction,” or to use a less expensive model for these background actions. See the full breakdown.

Whether you self-host or use the managed platform, your workspace, memories, and credentials belong to you. On the managed platform, your assistant runs on infrastructure provisioned exclusively for you. Conversations are sent to an AI model provider like Anthropic to generate responses, but are never stored or used for training. We’re transparent about that trade-off.

No. Your data is never used for training or fine-tuning from our side. For the AI model provider’s data policies, we recommend reviewing their terms directly.

Email, calendar, Slack, web browsing, food delivery, image generation, coding, app building, document writing, task management, phone calls, and more. Your assistant ships with 60+ skills out of the box, and you can teach them new ones to extend their capabilities further.

With your permission, yes. Your assistant can see your screen and interact with UI elements through macOS accessibility and screen recording permissions. Both are opt-in and require explicit system-level approval before they’re enabled.

macOS and iPhone are supported today. Web access is available from any browser with no download. Android and Windows are on the roadmap.