Vellum
Open-source personal AI assistant with persistent memory
A personal AI assistant that remembers you and keeps working in the background. Worth it if you will watch the credit meter and trust it with real accounts.
Founders and operators who want one always-on assistant across Mac, iPhone, web, Slack, and Telegram, carrying context between your inbox, calendar, and notes.
It bills you while you sleep. Background housekeeping burns credits even when you are not chatting, and pricing has changed repeatedly all year.
Reach for it when you want a personal assistant you shape over weeks. Skip it if you need a fixed monthly bill or a shared team tool.
What you're actually getting
Vellum relaunched in May 2026 as a personal AI assistant. You name it, shape how it talks, and it follows you across web, Mac, iPhone, Slack, Telegram, and email.
The core is memory. Eight memory types plus a personal knowledge base let it remember your preferences, open loops, and past decisions between sessions, so you stop re-explaining yourself every week.
It runs in Vellum's hosted cloud, or on your own computer if you prefer. The code is open source, and your passwords sit in a locked vault the assistant itself cannot read.
Big congrats to the Vellum team on launching their personal intelligence assistant — an enterprise-grade, secure, hosted Claw that learns who you are and truly becomes your assistant across work and life. I've been using it for two weeks and have been pretty blown away.
Where it earns its keep
Its best work happens while you are away. It can triage an inbox overnight, draft replies, brief you on Slack and your calendar before you open your laptop, and file expenses from receipt photos.
Memory pays off on research and recurring workflows. Ask it to dig into a topic and it builds on what it already knows about your business, then packages the findings into a report or a draft.
A shipped Marketing skill pack handles campaign busywork, with sales and customer-success packs planned. Bring your own AI account keys and running it on your own machine drops the hosting fee to zero.
Took me 20 minutes to build after i found vellum's agent builder. Just told it what I wanted. These took under an hour each with vellum so i could actually test ideas and tweak them based on what actually happened.
Where it'll bite you
It costs money while idle. The official FAQ admits it: background housekeeping runs even when you are not chatting, and that work consumes credits. You can turn the heartbeat down and route easy tasks to cheaper models, but budget for it.
Pricing has churned all year. Third-party sources captured entry prices of $15, $30, $50, and higher across 2026. Budget against today's pricing page, never against a review you read last quarter.
Real value means real access. You will connect your inbox, calendar, and other sensitive accounts, and grant permission to act on your machine. Access is gated from strict to full, and setup takes about three days to click. Treat it as a working relationship you maintain.
I think cost is the downside for vellum, it has everything else going for it.
What it costs, really
Entry is Mighty at $30/mo including $25 of credits. Super is $100/mo with $45 credits plus an assistant email address, and Ultra is $200/mo with $115 credits.
Past the included credits, usage is pay-as-you-go at $1 per credit, covering AI thinking, web searches, images, and paid third-party services. Your two levers are background activity and model choice: lower the heartbeat, send easy jobs to cheaper models. Running it on your own hardware removes the plan fee entirely, though you still pay your AI provider for usage.
Vellum works well for local permission controlled workflows. Lindy is more polished but cloud based and gets expensive with heavy use.
Nice ad 👌 saw your launch video on Twitter. All the best.
Same job, other tools
Also filed under Workflow & agent infrastructure:
