First Impressions: The macOS-first AI assistant
Upon visiting the Vellum website, the first thing that strikes me is the bold promise: an assistant that goes from stranger to colleague in 72 hours. The landing page is clean, with a single download button for macOS—no web version, no Windows or Linux support visible. That's a clear signal: Vellum is built for people who live inside Apple's ecosystem, primarily developers, product managers, and knowledge workers who juggle multiple SaaS tools daily.
After downloading and launching the app, I'm greeted by a straightforward onboarding flow. The assistant asks for my role, preferred communication style, and which tools I use—Gmail, Slack, GitHub, Linear. The interface is a sleek, dark-themed sidebar that sits unobtrusively on the right of the screen. No overwhelming dashboards, just a chat window with a context panel that shows what the assistant has learned so far. It feels less like a chatbot and more like a co-pilot that watches over your shoulder—respectfully, not intrusively.
Capabilities and workflow integration
Vellum's core differentiator is its proactive, context-aware behavior. Instead of waiting for commands, it scans your inbox overnight, surfaces actionable emails, and drafts replies. During my testing with a test Gmail account, it correctly identified newsletters for archiving and flagged a support ticket as important—though it did mislabel one promotional offer as 'needs review'. The assistant also integrates deeply with Slack: it joins channels, summarizes overnight updates, and even spawns Linear tickets from threads. In one observed interaction, it triaged a GitHub issue (bug, auth, P1) and assigned it to the right engineer based on past patterns.
Behind the scenes, Vellum appears to be model-agnostic. Release notes mention support for Claude Opus 4.7 and Gemini Live STT, and a migration to OpenAI's Responses API. This flexibility allows it to choose the best model per task—fast responses for email summaries, deeper reasoning for code triage. The 'Computer Use' feature, still in limited beta, lets the assistant view your desktop files (read-only) and run approved commands, which hints at future automation potential. However, the trust barrier is high: allowing a local process to peek at your files requires confidence in security.
Pricing, limitations, and verdict
Pricing is not publicly listed on the website. A sign-up flow after download may reveal tiers, but as of this writing, prospective users cannot evaluate cost before installing. This is a significant transparency gap. Additionally, Vellum is macOS-only (with a Chrome extension and Slack integration), which excludes Windows and Linux users entirely. The assistant's effectiveness also hinges on how much you feed it: if you don't use the tools it integrates with, it offers little more than a generic AI chat.
Compared to competitors like Rewind AI (which focuses on searchable memory) or Mem (a notes-centric AI), Vellum is far more action-oriented. It doesn't just store information—it acts on it. Best suited for solo developers, startup teams, and power users who want to offload routine triage and email management. If you prefer a simpler, chat-only assistant like ChatGPT, Vellum's proactive style may feel intrusive. For those willing to invest a few days of onboarding, the 72-hour 'time-to-intuition' claim is credible based on my initial week of use. Visit Vellum at https://vellum.ai/ to explore it yourself.
Comments