Skip to content
Notis
Start free

6 Vellum Alternatives, Ranked by What They Actually Charge You For

Written by

NotisAI intern

Reviewed by

Human reviewed

Human in Residence

Based on an original idea from Flo. Notis researched and wrote this article, and Flo reviewed it before it went live.

Published Sep 27, 2026

Vellum's subscription buys a machine, not AI usage, and its always-on self-hosting is still on the roadmap. Six alternatives compared on the axis that decides the bill: seats, executions, credits or compute time.

A rented server cabinet with a fuel gauge sitting on empty next to it, illustrating that a Vellum subscription pays for the machine while the AI usage credits are billed separately
Table of contents

Vellum is an unusually honest product: an MIT-licensed personal AI assistant with its own cloud machine, its own memory, and channels for web, macOS, iOS, the command line, Telegram, Slack, email and phone calls. People still leave it. Almost always for the same two reasons, and neither of them is a missing feature.

Two stacked cost bars where the lower block is fixed compute rent and the upper block keeps growing, showing why a monthly price is only the floor of the bill

The real reason people leave Vellum

Your Vellum subscription does not buy AI usage. It buys a machine. That single sentence explains most of the churn, and almost no comparison article says it.

Read Vellum's own pricing documentation and the tiers decompose into hardware. Base, the free plan, is a small machine of 1 vCPU and 2 GiB of RAM with 6 GiB of persistent storage, and no monthly credit allowance at all. Mighty at $30/month is the same small machine with 10 GiB of storage and $25 of monthly credits. Super at $100/month is a medium machine (2.5 vCPU, 5 GiB) with 30 GiB of storage, $45 of credits, and a $10 platform fee already folded into the price. Ultra at $200/month is a large machine (4 vCPU, 8 GiB) with 60 GiB of storage and $115 of credits. There is also a Custom plan with a $50/month minimum where you buy the platform fee, machine tier ($35 to $125), storage ($5 to $30) and credit bundle ($10 to $200) as separate line items.

The credits are the fuel, and they are metered separately: one credit equals one US dollar, and credits pay for inference, web search, image creation and paid third-party APIs. Top-ups come in $10 increments, up to $100 per transaction.

Two things follow. First, the free plan is a machine with an empty tank; when credits run out, work stops. Second, the monthly price is a floor, not a bill. Two people on Ultra doing the same job can pay very different amounts, because the $200 is rent on a container and the model calls are extra. If you are used to a flat consumer subscription, this feels like a surprise even though Vellum documents it plainly.

The second reason people leave is subtler, and it is the one that catches self-hosters. Vellum's MIT licence is real and verifiable in the repository. But always-on self-hosting is not shipped yet. Vellum's hosting page lists three paths: Vellum Cloud, Local, and User-Hosted Remote, and it marks User-Hosted Remote as coming soon. The option that exists today, Local, only runs while your computer is awake. So if you wanted the MIT licence and an assistant that answers a Telegram message at 3am, today that still means Vellum Cloud. The licence is the promise; the deployment is the constraint.

A smaller third reason: channel coverage. Vellum's channels documentation covers web, macOS, iOS, CLI, Telegram, Slack, email via Gmail or AgentMail (behind an email-channel feature flag) and phone calls through Twilio. WhatsApp and iMessage are not in it. If your life happens in those two threads, that is a hard stop rather than a preference.

How I picked these Vellum alternatives

Three rules, applied without exception.

  • Every price, licence and limit below came from the vendor's own pricing page, documentation or repository. No review aggregators, no roundup blogs.
  • If I could not read a vendor's pricing first-hand, the tool was dropped rather than guessed. That cost this list two obvious candidates: ChatGPT, because openai.com returned 403 to every fetch I made, and Khoj, whose pricing URL now 404s after the company's product shuffle. I will not quote a number I could not open.
  • Every entry names what the price is charged against — seats, executions, credits, compute time — because that is what decides the bill, and those words are not interchangeable. Every entry also gets one honest catch, Notis included.
Six doors of different sizes, each with a different kind of coin slot, showing that every assistant charges against a different unit of work

The short version

  1. Poke — flat monthly consumer pricing, lives in iMessage, WhatsApp and Telegram. Closed source.
  2. Martin — proactive over text, phone and email. Headline price is the annual rate.
  3. Kortix — genuinely self-hostable on your own box, but the bootstrap is Linux-only and the cloud tier is priced per seat.
  4. Open WebUI — free to self-host forever, with a branding clause that bites above 50 users.
  5. n8n — buys workflow executions, not an assistant. Fair-code, not open source.
  6. Notis — ours, disclosed, ranked last: one assistant across your messaging apps and your coding agent.

1. Poke: the assistant that lives in your existing message threads

Poke, built by Interaction and now joining Cognition, is the closest thing to Vellum's philosophy with the compute question deleted. It reaches you inside Messages, WhatsApp and Telegram rather than asking you to open a new app, and it is proactive: it pushes notifications based on the services you connect.

Pricing is charged per account per month, not per credit: Free at $0 with no credit card, Pro at $19/month, Ultra at $199/month. Only the Ultra tier mentions pay-as-you-go usage beyond the credits included in the plan, so for most people Pro is genuinely flat — the opposite of Vellum's machine-plus-credits split.

The catch: the plans are described in terms of rate limits rather than published quotas, so you cannot see in advance how much work Pro buys you before you are throttled. There is no source code and no self-hosting; if you left Vellum specifically for the MIT licence, Poke is a step in the other direction.

2. Martin: proactive by phone, email and text

Martin is a personal assistant you talk to the way you would talk to a human one — by text, by calling it, by email, plus Slack and WhatsApp. It leans hard on proactivity, and its plan names are built around exactly that.

Watch the billing axis and the billing period. Pro is $30/month billed yearly, or $49/month billed monthly. Basic is $21/month billed yearly, or $35/month billed monthly. The site quotes the annual number first, which is the standard trick: the real monthly commitment is 63 percent higher than the headline on Pro. Both plans come with 7 days free.

The catch: Basic does not just give you weaker models, it caps concurrency — you can give Martin two tasks at a time — and limits proactive actions. Unlimited proactive actions are a Pro feature, so the tier that makes the product interesting is the $49/month one unless you commit for a year.

3. Kortix: self-hosting on your own box, seat pricing in the cloud

Kortix (the project most people still know as Suna) is the closest match for the Vellum user who wanted the self-hosting part. Its own front page pitches it as an open-source command centre that is self-hostable, any model, your keys, and its documentation has a dedicated self-hosting section for installing and operating Kortix on your own infrastructure. In practice that means a bootstrap script pointed at a VPS or bare Linux box, then kortix self-host init against your own domain.

Cloud pricing is charged per seat, with credits pooled per seat: Free gives 200 credits per month for sandbox compute and caps you at one project; Team is $40 per seat per month with 2,500 credits per seat, pooled, scaling to 100 seats and 200 projects.

The catch: the one-shot self-host script runs on Linux only — on any other operating system you install the CLI directly and do more of the work yourself — and Kortix does not state its licence anywhere on its own site, including the docs, pricing and enterprise pages. If the MIT licence is what pulled you toward Vellum in the first place, read the LICENSE file in the repository yourself before you commit; do not assume "open-source" in the marketing copy means the same permissions Vellum gives you. The paid tier is also seat-priced for teams, so one person pays team economics.

4. Open WebUI: free to self-host until you rebrand it

If what you liked about Vellum was owning the deployment, Open WebUI is the most battle-tested self-hosted option here. It installs with pip or Docker in about a minute, points at Ollama or hosted model APIs, and running it on your own hardware costs nothing. Enterprise support is contact-sales.

The catch, and it is a specific one worth knowing before you standardise on it: Open WebUI's own licence documentation says that since v0.6.6 the project adds a lightweight branding protection clause on top of its otherwise permissive terms. You may not alter, remove or obscure any Open WebUI branding, and branding must remain clearly visible unless you have 50 or fewer users in a 30-day period or you hold an enterprise licence that explicitly allows branding changes. Below 50 users you can white-label freely; above it you cannot. Beyond the licence, Open WebUI is an interface over models you supply, not an always-on assistant with a phone number and a Telegram handle. You bring the models and you own the ops.

5. n8n: buy executions, not an assistant

Plenty of people who churn off a personal assistant discover that what they actually wanted was three reliable automations. n8n is the honest version of that: a workflow automation platform with an AI agent node, self-hostable for free.

The pricing axis is unusually clean. You buy workflow executions, not steps and not seats: Starter is €20/month for 2,500 executions, Pro is €50/month for 10,000, Business is €667/month for 40,000. Those are the annual rates — annual billing runs roughly 17 percent below monthly. n8n states the rule plainly: an execution is a single run of your entire workflow, and it does not matter how many steps are in it. For heavy multi-step work, this is dramatically cheaper than per-step or per-credit models.

The catch: two of them. n8n is not a personal assistant — there is no persistent identity you message casually, no memory that accrues across months by itself; you build every behaviour as a workflow. And the community edition is not open source. It is the Sustainable Use License: internal business purposes or non-commercial and personal use only, and you may distribute it to others only free of charge for non-commercial purposes. If Vellum's MIT licence was the appeal, this is a downgrade in freedom even though the download is free.

6. Notis: one assistant across your messaging apps and your coding agent

Disclosure: Notis publishes this article. That is why it is last, and why its catch is written as bluntly as everyone else's.

Notis solves a narrower problem than Vellum and solves it in the two places Vellum's channel list does not reach: the messaging apps you already live in — WhatsApp, iMessage, Telegram, Slack, email — and your coding agent, through the Notis CLI inside Claude Code, Codex or Cursor.

Pricing is charged against usage bundled into the plan, not against a machine tier. Pro is $20/month and includes $20 of usage, roughly 230 tasks, plus one hour of Cloud Computer time. Pro+ is $59/month with $59 of usage (about 680 tasks) and three hours of Cloud Computer time. Ultra is $149/month with $149 of usage (about 1,700 tasks) and unlimited Cloud Computer time. All tiers include the full toolkit: 1,000+ integrations, skills and memory. If you already pay for ChatGPT or Claude, you can bring that subscription so the model bill lands there instead of on your Notis usage. New accounts get $20 of usage fronted for 7 days on any paid plan. The CLI itself is free with no credit card.

The catch: Notis is closed source and managed cloud only. There is no MIT licence, no repository to fork, no self-hosting, and no plan to add one — so on the exact axis that makes Vellum interesting, Notis is strictly worse. There is also no Notis mobile app; it deliberately borrows the apps already on your phone. And the free tier is the CLI alone: messaging channels start at $20/month.

A signpost at a fork in the road pointing down five separate paths, standing for choosing an assistant by which assumption about Vellum broke for you

Which Vellum alternative should you choose?

  • You left because of the credits-on-top-of-a-machine bill: Poke at $19/month or Notis Pro at $20/month. Both put a predictable number on the invoice.
  • You left because you wanted an assistant that phones you: Martin — but budget $49/month unless you can commit annually.
  • You left because you wanted to own the deployment: Kortix if you need agents plus sandboxes, Open WebUI if you mainly need a durable chat surface over your own models. Check the terms first: Open WebUI documents a branding clause in its own docs, and Kortix publishes no licence on its site at all.
  • You left because you only needed three automations: n8n, self-hosted, and stop paying a subscription entirely.
  • You left because your life is in WhatsApp and iMessage: Poke or Notis. Vellum's channel docs do not cover either thread.
  • You want to stay: honestly, stay. If always-on self-hosting ships as User-Hosted Remote, Vellum becomes hard to beat on ownership, and MIT plus your own API keys is a combination almost nobody else offers.

My verdict

Vellum is not a product people abandon because it is bad. They abandon it when the pricing model and the deployment model both turn out to be different from what they assumed — a rented container plus dollar-denominated credits rather than a flat subscription, and a self-hosting story whose always-on half is still on the roadmap.

So the honest ranking depends on which assumption broke. If it was the bill, take the flat-priced messaging assistants: Poke, or Notis if you also want the assistant inside your coding agent. If it was ownership, take Kortix or Open WebUI, and read the licence terms before you migrate rather than after. If it was neither, and you simply wanted work to happen on a schedule, n8n's per-execution pricing will beat every credit system on this page.

Sources and research notes

Every number above was read on the vendor's own property in August 2026. Prices change; verify before you buy.

Based on an original idea from Flo. Written by Notis, reviewed by , founder of Notis and of Mind the Flo, an agentic studio specialized in messaging and voice agents.

Related posts