The example on the Notis website is clearing an inbox. In the illustration, 14 unread messages come in, and it comes back with: 11 archived, 3 need you, two replies drafted. A heavier job is “compare the 4 CRMs on my list and give me a document,” and it delivers an 8-page comparison saved to Drive. (Notis pricing[1])
The page adds a qualifier of its own: these examples only show what you can do; the product is not priced per task. (Notis pricing[1])
I remembered the word “drafting.” It did not say it would send it for you.
Image: Notis official showcase[2], retrieved 2026-11-10, original interface retained.
The second arrangement looks more like covering a shift for someone: when the Mac is on, Notis runs your agent on your own machine; when the Mac is asleep, it switches the work to a machine in the cloud to continue. It chooses which machine automatically, and you can also pin a specific one for a channel, automation, or conversation. (Notis pricing[1]) Long tasks can keep running in the background, and when they finish it pings you in the same thread; the chat box does not need to be open. (Cloud Computer[3])
That cloud computer, and the places it will not guess
What holds up “work can continue when the person is not there” is Cloud Computer: a private cloud workspace where running scripts, opening a browser, editing files, and writing full documents do not touch the user’s computer. Files, installed command-line tools, and login state stay in place for next time, and every paid plan includes it. Skills in the free plan still sync everywhere, but they actually run on the user’s own machine, through the desktop app or command line. (Cloud Computer[3])
One detail is worth pausing on: it will install command-line tools, npm packages, and Python libraries while running, but when credentials or permissions are missing, it asks the user to connect them rather than guessing. (Cloud Computer[3]) “Not guessing” draws a line of responsibility—if it guesses one account wrong, the user bears the cost.
Money takes two paths
One is models. Connect a ChatGPT or Claude subscription you are already paying for, and that agent’s model tokens go through the original subscription, not against Notis’s usage allowance. (Notis pricing[1])
The other is doing the work. In-progress cloud tasks, together with models, paid tools, and voice, consume that monthly Notis allowance; there is no separate “hourly package.” (Notis pricing[1]) Model usage called through Notis, paid tools, voice, and in-progress cloud work are billed at twice the provider cost; model tokens connected through your own subscription do not go through this ledger. Reminders, email, research, documents, automation, and social tasks share the same allowance; optional overage is settled after the fact within a user-set cap. (Notis pricing[1]) Pro includes $20 of usage per month, Pro+ includes $59 of usage per month, and Ultra includes $149 of usage per month. (Notis pricing[1])
For users who connect their own subscription, the model bill stays where it is, and Notis only charges for the work side. The cost is that users have no per-task price list: clearing an inbox once and producing an eight-page document both draw from the same pool, and there is no way to know in advance how much each will cost.
One person, and one burnout
Flo’s company has only one person, and the product is growing into a shape that “one person can manage.”
Flo’s founder page puts it plainly: he built Notis alone for two years, and round-the-clock intense focus brought him to burnout and one ADHD diagnosis, after which he rebuilt the product around the help he needed. (About Flo[4])
Notis was not like this at first either. It began as one WhatsApp-to-Notion capture workflow; later, based on actual usage, it expanded from that light action into an assistant and automation platform. (Notis blog[5])
He now runs the company by messaging the product: setting direction, talking with users, reviewing results one by one, handing off research, content, follow-up, and operations work, and using a coding agent to help deliver code. (About Flo[4])
There is a line here that must not be blurred. “Reviewing results one by one” is his human attention; it is not the same as the approval rules in the product—public materials do not say whether Notis has set product-level approval for external actions. Compare Lindy’s approval rules: its pricing page states explicitly that actions with external impact, such as sending email, updating tickets, posting to other channels, and publishing documents, wait for approval, while read-only queries from approved sources do not. (Lindy pricing[6]) One person keeping an eye on every item, and a set of rules keeping watch for everyone, are two different businesses.
The ledger, laid open
In a snapshot from November 14, 2026, 02:12, TrustMRR records Notis’s plan subscription prices as Free $0, with Pro, Pro+, and ULTRA at $13, $39, and $99 per month respectively, all three tiers counted on an annual-payment basis; the same snapshot shows cumulative revenue of $244,464, MRR of $13,135, and 426 active subscriptions, with data verified through a Stripe API key. (TrustMRR[7])
The boundary needs to be clear: MRR is not profit, and active subscriptions are not customer count. So how much this business earns is invisible from the outside; all that can be seen is that it is indeed collecting money.
Flo’s own page says “funding $0,” bootstrapped, profitable, founder-run, and compares it with Viktor’s $75 million Series A led by Accel. (About Flo[4]) The blog post dated March 19, 2026, is bylined by Notis’s AI intern and reviewed by Flo; it says he is not against VC but against the default option, and if he did raise money, it would only be to hire a very small number of AI-native people. (Notis blog[5])
There is also something not on the plan table: Flo-as-a-service, one hour per week, starting with a 30-minute discovery call, with limited slots, and the reason is stated plainly—the company has only one person. (About Flo[4]) The page does not list a price. It is a publicly listed offering.
The bet Notis is making is not “AI can do the work,” but leaving the fallback position in the hands of one specific person: users connect missing credentials themselves, users set the overage cap themselves, and the task examples say “draft” replies. For now, this arrangement is still held up by one person. If it grows further, it will first have to answer: how do product rules catch actions with external impact? Public materials do not yet have that answer.