Vellum

The Short Version

Vellum is a memory-first personal AI assistant from Vellum AI. Instead of answering one-off prompts and forgetting everything, it builds a persistent memory of your preferences and context, then acts across the channels you already use. You reach one assistant, with one shared memory, from a macOS app, Slack, Telegram, and other channels. It’s aimed at people who want an assistant that behaves less like a chatbot and more like a teammate.

It is also open source under the MIT license, so you can inspect how it works and run it yourself.

What Vellum Is Best For

• Personal productivity: an assistant that remembers your preferences and working style
• Reducing repeated context: stop re-explaining the same background every session
• Cross-channel work: reach the same assistant and memory from macOS, Slack, or Telegram
• Task handling: delegate recurring work rather than only asking questions
• Privacy-conscious users: self-host under an open license if you’d rather not use a managed cloud

How Vellum Works

Vellum extracts memory from your conversations, actions, and notes rather than only from what you explicitly write down. That memory is structured, with a small always-loaded working set plus a larger store it retrieves from as needed, so context carries across sessions and channels. On top of that memory, it can connect to your accounts through managed OAuth and take actions like triaging email or managing your calendar, with permission gating on sensitive operations. It also runs proactively, re-checking its notes on a schedule and flagging anything that looks unfinished or due.

You can run it two ways: managed, by signing in through Vellum Cloud, or locally, where everything runs on your own machine.

Practical Use Cases

• Remember preferences and context: keep your working style available to the assistant over time
• Coordinate work in Slack: reach the assistant where your team already talks
• Handle recurring tasks: hand off routine work instead of redoing it manually
• Act as a personal work assistant: email triage, calendar help, and follow-ups
• Keep one memory everywhere: switch channels without losing context
• Self-host for control: run it on your own machine when privacy matters

A Note on Privacy

Because Vellum’s value comes from remembering you and connecting to your tools, treat setup carefully. Review what it stores, which accounts and apps it can reach, and where that data lives. Avoid feeding it credentials, secrets, or sensitive company data without a proper security review, and lean on the self-hosted option if you want to keep everything local. Memory-based assistants trade some privacy for convenience, so make that trade deliberately.

Honest Limitations

• Trust and privacy review required: a memory-first assistant needs careful setup before sensitive use
• Setup effort: self-hosting and connecting accounts takes work, especially the local option
• Costs beyond the app: even the open-source version has model and compute costs
• Verify capabilities: features and pricing change, so confirm the current state on the official site

Alternatives Worth Knowing

• ChatGPT, a general assistant with its own memory features
• Claude, strong for reasoning and long-context work
• Mem, an AI note tool built around memory
• Notion, a workspace with built-in AI
• Lindy, an AI assistant focused on automating tasks
• Relay, an AI assistant for coordinating work