The Kiwi proactive AI assistant does something most assistants never do: it texts you first. Instead of waiting for you to open an app and ask, Kiwi watches your calendar and email, spots the small job that needs doing, and sends you one message at the right moment. You reply yes, and it is done. Our verdict up front: for busy people drowning in small admin, this is the most interesting assistant design to launch in months — but it is still a pilot, so treat it as a promising experiment, not as infrastructure you bet your business on yet.
Kiwi launched on Product Hunt on 24 October 2026, built by maker Aneesh, and its pitch is deliberately simple: a personal assistant costs about $67,000 a year, and Kiwi does the same small jobs every week for nothing — at least while the pilot lasts. There is no app you must live in. You talk to Kiwi in iMessage or WhatsApp, on iPhone or Android, and setup starts with a pre-written “Hi Kiwi” message. That choice matters more than it looks: the assistant meets you where your attention already is, instead of asking you to build a new habit around yet another app.
What Kiwi Actually Does
Kiwi reads your calendar and email first, then adds the apps you already use — Google Calendar, Gmail and Outlook to start, then Slack, WhatsApp, Spotify, Uber and more. When there is no app for a job, it uses the website directly. From that picture of your life it builds three kinds of help: a short morning brief of what is coming, a single well-timed text when something needs doing, and action on your yes.
The examples on Kiwi’s official site show the pattern clearly. Your Saturday afternoon is free, so Kiwi asks whether you want it to plan something with the kids — and if you say yes, it books the tickets, lunch and the taxi. Your flight is delayed an hour, so Kiwi offers to book a taxi from arrivals for your new landing time. A parcel needs a signature while you are at work until 6, so Kiwi offers to move delivery to Saturday — and only reports it done once the carrier confirms. A flight to Rome you asked it to watch drops from €112 to €84, and Kiwi texts you with the link. Dinner is at 9, so at 6:30 pm it offers a car for 8:50, with both prices shown up front: an UberX at about €14 or a Comfort at €18.
Three design rules run through all of this. First, anything that costs money shows its price before you commit. Second, Kiwi only says something is booked once the company confirms it, and it never pays twice for the same thing. Third, after each conversation it goes back over what happened and fixes anything it got wrong or forgot. That last habit — a self-review pass — is the kind of unglamorous reliability feature that separates a demo from a tool you can trust with real bookings.
Pricing: Free During the Pilot, and What That Really Means
Kiwi is free during the pilot. There is no card to enter, no plan to pick, no ads, and the company states plainly that your data is never sold and no business can pay to change what Kiwi suggests. Its comparison is pointed: the average base pay of a US executive assistant in 2026 is $67,567 a year according to Payscale, or $1,299 a week — and Kiwi’s receipt for an example week totals $0.00.
Be clear-eyed about what “free during the pilot” means. It is a launch offer, not a business model, and Kiwi has not announced what pricing looks like after the pilot. That is normal for a product at this stage, but it changes how you should evaluate it. Use the pilot to test whether proactive assistance genuinely saves you time. Do not build a business process that depends on Kiwi staying free, and keep your own records of anything it books, so a future price change never leaves you stranded. If and when paid tiers appear, judge them on the same question as any tool: does the time saved beat the subscription, measured in your real week, not the marketing example?
Who Kiwi Is For (and Who Should Wait)
Kiwi fits people whose days are full of small, time-sensitive jobs: frequent travellers, parents juggling family logistics, founders and managers whose calendars fill faster than they can triage them. If your pain is not “I cannot think of what to do” but “I know what to do and never get around to the ten small steps”, a proactive assistant is aimed exactly at you. The morning brief plus one timely text is a genuinely different shape from a chatbot you must remember to consult.
Who should wait? Anyone handling regulated or highly sensitive data, anyone whose bookings are high-value and irreversible without a human check, and anyone uncomfortable giving an AI agent read access to their email and calendar. That access is the whole product — without it, Kiwi cannot know what needs doing — so if that trade is a no for you, no feature list will fix it. Teams should also wait for clearer multi-user, permissions and audit features before letting an assistant act across shared inboxes or company accounts. And if you need a tool with published pricing and a service-level commitment today, a pilot is the wrong place to look.
Watch-Outs and Honest Limitations
The biggest risk with any proactive assistant is notification overload — the exact problem it claims to solve. Kiwi’s answer is discipline: one text at the right time, with anything that can wait held for the morning summary. That discipline is easy to promise and hard to keep as the product grows. If you try Kiwi, judge it on week two, not day one: are the texts still welcome, or have they become one more feed to manage? The maker is candid about rough edges too. On the Kiwi Product Hunt page, he notes that some sites ask you to prove you are human, and invites users to report where it breaks. That is honest, and it is also a real limitation: any agent that drives websites will hit CAPTCHAs, login walls and anti-bot checks, and those failures land on your errand, not on the demo.
Security deserves a careful read. Kiwi states that passwords are encrypted with AES-256-GCM before they are saved, that the AI never sees them, that nobody at Kiwi reads your conversations, and that you can delete a single memory or your whole account and everything in it. It also states that nothing written in an email or on a website can tell Kiwi what to do — only you can. That last point is a direct answer to prompt-injection attacks, where a malicious email tries to steer an agent. These are strong claims for a young product, and they are exactly the claims you should verify before handing over credentials: read the data-handling page, test the delete controls, and start with low-stakes accounts before connecting anything critical.
Finally, remember the pilot caveat in every direction: features, pricing, availability and even the company itself can change. Kiwi’s wristband, the Kiwi Band, is a good example — it is prominently advertised, but it is explicitly “coming later” and not available yet. Judge the product on what ships today, not on the roadmap.
How Kiwi Compares to the Other Text-First Assistants
Kiwi is not alone in the “text an assistant, it does the job” space, and the differences are instructive. Tend, another recent Product Hunt launch, is also text-first through iMessage and SMS, but it leans business: it connects Gmail, calendar, Shopify, QuickBooks and HubSpot, keeps what it learns in a readable “Brain” you can correct, and is invite-only during early access. Hark Pro takes a different route — a personal AI with its own cloud computer, a privacy pitch built on encrypted credentials and no data selling, and the ability to watch the agent while it works. Atticus is the outlier: a free, open-source Mac assistant that runs simple asks instantly without AI and borrows your existing Claude plan for the rest.
Kiwi’s differentiator is proactivity as the default, not a feature. Tend and Hark both offer proactive nudges, but their centre of gravity is still “text it a task”. Kiwi’s centre of gravity is “it texts you”. Its other differentiator is phone-native design: no new app to live in, no desktop required, and a screenless band on the roadmap that would push the whole idea further — your phone stays in your pocket, and the assistant reaches you with a buzz, a tap and your voice. The trade-off is scope: a phone-first assistant that works through messages will not match a desktop agent with a cloud computer for long, multi-step computer work. Pick by the shape of your day, not by the feature table.
From Our Experience: Where Proactive Assistants Win and Where They Break
At MTD Technologies we have spent 9+ years building web, mobile and AI systems for clients, and the pattern with automation is consistent: the tools that survive contact with real users are the ones with a human confirmation step in exactly the right place. Fully autonomous agents sound impressive and fail quietly — a wrong booking discovered three days later costs more than the ten minutes it saved. Kiwi’s “say yes and it is done” model gets this right: the human decision is one word, but it is still a decision, and price-first plus confirmation-only reporting are the two guardrails that matter most. If you are evaluating any assistant for your business, copy those two checks into your requirements list.
The second lesson is about notification tuning. Every proactive system we have seen starts delightful and drifts noisy, because each new integration adds a new reason to ping you. The fix is not fewer features but better thresholds: only interrupt for things that get worse with waiting — a fare that will rise, a slot that will fill, a deadline that will pass — and batch everything else into a digest. Kiwi’s morning summary is the right instinct. When you trial it, keep a simple scorecard for two weeks: how many texts arrived, how many you acted on, and how many you wish had waited. If the action rate falls below half, the assistant is costing you attention, not saving it.
And a word on when NOT to use a proactive assistant for business work: never let one act on a shared inbox, a finance mailbox or a client-facing account without an audit trail you control. Kiwi’s delete controls and “only you can instruct me” rule are good starting points, but for client work you also need logs, role separation and a way to revoke access instantly. That is the gap between a great personal tool and a business system — and it is exactly the kind of gap our AI automation services team closes when we build agent workflows for clients. Browse the AI Tools Directory for more tools we have reviewed in this series, and if you want help deciding which assistant fits your workflow, talk to us — we will tell you honestly when the answer is “none of them yet”.
Frequently Asked Questions
Is Kiwi really free?
Yes, during the pilot. There is no card to enter and no plan to pick, and Kiwi states there are no ads, your data is never sold, and no business can pay to change what it suggests. Post-pilot pricing has not been announced, so treat the pilot as a trial window and keep your own records of bookings.
How does Kiwi know what needs doing?
It reads your calendar and email, then watches the apps you connect. Things that can wait go into a morning summary; things that need action now arrive as a single text at the right moment, with the small job already worked out so your reply can be one word.
Can Kiwi spend my money without asking?
No. Anything that costs money shows its price first, and nothing is paid or sent without your yes. Kiwi only reports something as booked once the company confirms it, and it never pays twice for the same thing.
Is my data safe with Kiwi?
Kiwi states that passwords are encrypted with AES-256-GCM and never seen by the AI, that nobody at Kiwi reads your conversations, that nothing in an email or on a website can instruct it, and that you can delete a single memory or your entire account. Verify the data-handling page and start with low-stakes accounts before connecting anything critical.
What is the Kiwi Band?
A screenless band for your wrist, coming later and not available yet. It buzzes when Kiwi needs you, and you answer with a tap or your voice, so your phone can stay in your pocket. It is the most interesting part of Kiwi’s vision — and the part you cannot buy today, so do not factor it into a decision now.
The Kiwi proactive AI assistant is worth a two-week trial if your days are full of small, time-sensitive jobs and you are comfortable with calendar and email access in exchange for fewer of them. Start small: one mailbox, one calendar, low-stakes errands, and a scorecard of every text it sends. If the action rate stays high and the texts stay welcome, you have found a genuinely new shape of assistant. If not, you have spent nothing and learned exactly what to demand from the next one — and that is a useful result either way.