Town deep dive: the personal AI assistant with a bunny on its shoulder.

Town is a personal AI assistant you name and give a shape to — the founders call the resulting character a ‘Townie’ — then connect to the surfaces where your day actually happens: email, calendar, Slack, Google Drive, Zoom, Granola notes. Once connected, it reads your context, suggests things you could do next, drafts the replies and briefs, and waits for you to press send. That is the whole product, and it is a lot more product than it sounds.

The company came out of beta at the start of June 2026 and its users have more than quadrupled since. It closed a $55 million Series A led by Justine Moore at Andreessen Horowitz in the same month. The two co-founders are Jean-Denis Greze, formerly chief technology officer at Plaid, and Tony Vincent, formerly director of applied AI Product at Google — they overlapped at Dropbox roughly a decade ago. This is the deep dive: what it is, what it does end-to-end, what it costs, where it fits in a health stack, and where a careful user should keep a hand on the wheel.

what a Townie actually is

The Townie is the on-screen character — a bunny, a wombat, a pink axolotl, a silver fox, whatever you pick — and the personality wrapper around the underlying assistant. It is not cosmetic. The founders’ design bet is that giving the tool a name and a form makes it ‘sit in your mind differently,’ as Vincent put it, and lowers the threshold of trust required to hand it your inbox. The evidence from the reference piece is that this worked: the product spread through Andreessen Horowitz ‘like wildfire’ and through a Sydney plumbing shop at the same time, largely by word of mouth.

Underneath the wrapper, Town uses a blend of frontier and open-source AI models, according to the founders. The choice of model on any given task is Town’s to make. The user does not swap models by hand. That is a deliberate simplification — a bet that the boring part of intelligent-system design (routing the right question to the right model) is Town’s job, not yours.

what it does, end-to-end

Once you connect the surfaces, the loop looks like this.

1. Ingest — Town reads across your connected surfaces (inbox, calendar, chat, drive, meeting notes) and builds an ongoing model of what you are working on and who with.
2. Suggest — it comes back with a small, quiet menu of things it noticed you could do next: reply to this thread, prep for tomorrow’s call, follow up on that intro, book the exploratory chat.
3. Draft — when you say yes, it produces the actual artefact (the reply, the brief, the slide, the summary) in your voice, using context from every surface it has access to.
4. Approve — you review the draft and press send. Nothing leaves your account without that step.
5. Delegate — where you have a human assistant or teammate whose Townie is also in the system, your Townie can pass the finished thing to theirs to complete the loop.

The reference article gives good, concrete examples. Justine Moore’s Townie Clio automates her response to warm investor intros by drafting an email looping in her human assistant to schedule an exploratory call. Kirsten Green’s Townie Maxie preps her for meetings, drafts emails and reviews deals across 76 portfolio companies. At Forerunner Ventures, individual Townies talk to each other: Bourne (an owl) asked Cami (a fox) for a summary of June support work for portfolio companies, and Cami produced a rundown Kira McCroden described as ‘way more expansive than anything I would have captured myself.’

“The interesting technical claim is not that a Townie can draft an email. It is that two Townies can complete a loop end-to-end — and then hand the finished thing back to the two humans to gently review and send.”

pricing, in plain numbers

Individual subscriptions run from $15 to $199 a month, depending on how many credits you use. Team plans start at $59 per seat. Most of the people quoted in the Inc. piece were still on a free trial. That is a wide pricing band, which tells you two things: Town is metering compute usage carefully, and heavy users are absorbing meaningful costs. For a personal health use case where the assistant is producing a weekly brief and drafting a handful of messages, you would expect to sit near the low end. For a small practitioner team automating intake, follow-ups and roundups across several inboxes, expect the mid-band on a per-seat basis.

where it plausibly fits in a wellness stack

Nothing about Town is health-specific. That is a feature, not a bug — it means the same tool that runs your inbox can, with the right consent and the right folder structure, read the surfaces where your own health lives. The three fits worth naming:

• The weekly self-brief — give the Townie access to a single folder that contains your wearable exports, your journal notes and any recent lab PDFs. Ask for a two-sentence-per-surface Monday brief. Judge it on whether it makes your week quieter, not on whether it impressed you.
• The pre-appointment prep — before a GP, specialist or coach call, ask for a one-page summary of what has actually changed since the last visit, sourced only from the files in the folder. You bring it to the appointment; you do not send it to the clinician.
• The practitioner triage — for a solo or small-team practice, drafts of the repetitive replies (intake, rescheduling, follow-up, referral) with a strict human-at-the-edge rule for anything that touches clinical judgement.

None of these is exotic. All of them are the kind of loop most people currently do by hand at 10 pm on a Sunday. That is exactly the point — the win is not intellectual novelty, it is compressing the boring middle.

where it does not fit — and the honest risks

Town is a consumer product and should be treated as one. It is not a clinical tool, it is not a regulated medical device, and its answers are not medical advice. Do not use a Townie as a stand-in for a clinician who has actually seen your labs. The Bryce Davidson quote in the piece — the T-shirt printer who was ‘worried about Town knowing too much about him’ before he trusted it — is the correct instinct, not a solved problem.

The concrete risks worth naming out loud:

• Consent surface — you are granting a third-party service ingestion access to inbox, calendar, chat and files. Assume everything you connect is now readable by the assistant on your behalf, and only connect the surfaces you have actively decided to include.
• Voice capture — Town is unusually good at matching your writing style, which is exactly why every draft still needs a human read. A voice-matched wrong answer is worse than a stilted right one.
• Sycophancy trap — Davidson praised Town for being direct rather than flattering. Trust that only as long as it holds. When any assistant starts telling you what you want to hear, retire the routine.
• Model opacity — the blend of frontier and open-source models is Town’s choice, not yours. If you have a strict data-residency or model-transparency requirement (e.g. an EU practitioner with GDPR concerns), this may be a dealbreaker.
• Health specificity — general-purpose assistants average to the population. If you are the interesting case, insist on citations for anything health-related and treat the draft as a hypothesis, not a plan.

how it compares to what you already have

The honest comparison is with the assistant already built into your daily chat tool. That built-in assistant is free, reads the surfaces you paste in, and is perfectly adequate for the weekly-brief use case if you are willing to keep the folder tidy and the prompt template consistent. What Town gives you on top of that is passive ingestion — you stop having to paste — and the multi-agent loop where a Townie can hand a finished draft to a colleague’s Townie for the next step. That is worth real money if your day looks like Kirsten Green’s and much less obvious money if your day looks like a plumber’s Sunday inbox once a week.

The right framing is not ‘replace what I use with Town.’ It is ‘which of my weekly loops is boring enough, and boundaried enough, that I would happily delegate the middle of it to a named cartoon animal I have to sign off before it sends.’ That is a small list for most people. That is fine. A small list, run well, is the whole win.

verdict

Town is one of the first personal AI assistants that reads like it was built by people with taste, not just people with distribution. The wrapper is charming; the integrations are wide; the ‘human at the edge’ posture is exactly right for anything that touches a life outside a computer. It is not the tool that will read your labs for you — nothing on the market is — but it is a very good example of what the personal-assistant layer should feel like, and the habits it forces (one folder, weekly brief, quiet menu of suggestions) are the right ones to build even if you never install it.

Try it if your week is a Kirsten Green week. Steal the habits if it is not. Either way, the assistant layer just got real, and the taste question — whose draft is worth reading — is the question you should be asking of every tool you already use.