Gyrodile
Ads Creators Search Tech Insights
Schedule a call
← All insights
AI Automation 8 min read · September 2026 Tom Whitfield Tom Whitfield

KuraChat Review: The Honest Version, From the Team That Built It

A first-party review of KuraChat, the AI support widget we built for small businesses: what it does, how a message is actually handled, what flat pricing saves against…

SHARE 𝕏 in
KuraChat Review: The Honest Version, From the Team That Built It

Disclosure: we built KuraChat. This is a first-party review, so treat the praise with the scepticism it deserves and hold us to the section on what it does not do yet, which is the part most vendor pages leave out.

Small businesses answer the same fifty questions forever. What time do you open, do you ship to Ireland, where is my order, can I move my appointment. The answers exist, usually in a policy page nobody reads, and the questions still arrive at eleven at night when there is nobody to answer them.

KuraChat is our attempt at the smallest useful tool for that problem: an AI support widget that answers from your own content, plus a shared inbox so a human can take over when it matters. This review covers what it actually does, how a message is handled underneath, what it costs against the per resolution pricing that dominates this category, and the five things it does not do yet.

The core position

Support tools are priced for companies with support departments. KuraChat is built for the business where support is the owner, one person on Saturdays, and a phone. That constraint decided everything: flat pricing, one line of install, and an AI that is never allowed to pretend it is a person.

What it is, in one screen

A single snippet on your site puts a chat widget on every page. The assistant answers from knowledge you give it: business details, FAQs and policies as plain text, uploaded documents, and pages from your own site. When a customer asks for a person, or the assistant is not confident, the conversation moves to a shared inbox with the full transcript attached.

$0 Free: 100 conversations, 1 seat AI chatbot, FAQs, handoff, no card $29 Pro: 1,000 conversations, 3 seats Documents, integrations, analytics $99 Ultra: unlimited conversations Unlimited agents, SLA, priority support $0.00 charged per AI resolution, on every plan the category norm is about a dollar each

Setup, honestly timed

Installation is the part vendors exaggerate, so here is the real sequence. Create a workspace, fill in a short form of business details, paste your FAQs and policies as text, then copy one snippet into your site template. On WooCommerce there is a WordPress plugin instead, and on Shopify a one click OAuth install that injects the widget for you. That part genuinely takes minutes.

The part that takes longer, and that decides whether the thing is any good, is the knowledge. An assistant pointed at three sentences answers like it has read three sentences. Uploading your real documents (PDF, Word, Excel, CSV or text, on paid plans) and adding your site’s own pages as sources is what moves it from a novelty to something that deflects work. There is a test console in the dashboard that shows which knowledge entry produced each answer, which is the fastest way to find the gaps before a customer does.

Small shop owner at a counter in the evening with a laptop showing a chat interface

What happens to one message

Most chat widgets are a thin wrapper around a model call. The interesting design decisions are in what happens before and after that call, so here is the actual path a customer message takes.

How one message is handled Handoff scan a keyword match skips the model entirely Verification gate order lookups need a 6 digit email code Context assembly details, FAQs, documents, live store data Model reply last 15 messages, owner rules applied Sources, analysis answered from, summary, sentiment, confidence Two of the five stages exist to avoid calling the model at all, or to avoid answering with data the visitor has not proved is theirs.

Two of those stages are worth dwelling on. The handoff scan runs before any model call, so “can I speak to a human” is never answered by an AI being clever; it routes straight to the queue with your reply time and office hours. The verification gate is the one we are most pleased with: an order number alone gets a status and nothing else, and anything tied to a person’s account requires a six digit code sent to their email first. A support bot that will read out an order history to whoever types an email address is a data leak with a chat bubble on it.

Honest AI is not a setting

Every reply carries an AI label, the conversation opens with a one line disclosure you can reword but not remove, and if a customer asks, the assistant says plainly that it is an AI and offers the team. There is no stealth mode and no switch to add one.

That is partly conviction and partly the law. Article 50 of the EU AI Act requires that people are told when they are interacting with an AI system, and the obligation sits with the provider. Building the disclosure in as a default, rather than as a toggle a customer can quietly switch off, is the only version of this we would be comfortable shipping. It also removes the single most common reason chat conversations turn hostile: the moment a customer works out that the “agent” was never a person.

Empty hotel reception desk late at night lit by a single lamp

The pricing argument

This category has settled on per resolution pricing. Intercom’s Fin, the reference point everyone compares against, lists at $0.99 per resolution with a minimum of 50 a month. It is a defensible model: you pay when the machine does the work. It is also unpredictable in exactly the month you can least afford it, because a busy month is a bigger bill.

Run the arithmetic on a small store doing 1,000 conversations a month with 60 percent resolved by AI. That is 600 resolutions, about $594, before any seat costs. The same volume on KuraChat Pro is $29 with three seats included. The comparison is not entirely fair, because Intercom is a much larger product with a reporting and automation surface KuraChat does not have, and a company at real scale may want exactly that. For a hotel with four rooms of questions and one person answering them, paying per answer is the wrong shape of bill.

One honest caveat on our own side: paid checkout is not self serve yet. Upgrades are handled by us while the payment gateway work finishes, which is fine at our current size and would not be fine at ten times it.

Where it falls short

The list we would want to read if someone else had built this:

There is also a deliberate scaling ceiling. The whole thing is a Rails 8 monolith on SQLite with no Redis, which is why it is cheap to run and simple to keep alive, and why it is not the right choice for a support team of forty. The stack is the same server rendered Hotwire architecture we argued for in our piece on server-side rendering, and the judgement calls behind it are the ones we described in cross-platform development in a vibe coded world: humans own the data model, the authorisation and the security gates, whatever writes the screens.

Who it is for

KuraChat suits a business with one to three people covering support, a website that already explains most of what customers need, and a stream of repeat questions that do not require judgement: hotels and guest houses, small online shops, clinics and studios, service agencies collecting enquiries overnight. The WooCommerce and Shopify integrations make it noticeably more useful for stores, because “where is my order” stops being a human job.

It is the wrong tool if you need omnichannel coverage, advanced routing and workforce management, deep analytics, or a vendor with a procurement process to satisfy. Those products exist, they cost what they cost, and they are worth it at the size where support is a department rather than a person.

Verdict

The honest summary of our own product: KuraChat does one job properly, which is answering the fifty questions a small business is asked forever, in the business’s own voice, without pretending to be human, and handing over cleanly when a person is needed. It is priced so the bill does not punish a good month. It is missing self serve billing, background processing and any channel beyond the website, and we would rather say so than let you find out in week two.

If your support load is a handful of repeated questions and the person answering them has a business to run, the free plan costs nothing to test and a hundred conversations is enough to know. If you want a second opinion on where an AI assistant fits in your funnel before you install anything, that is what our Tech engine is for.

Try KuraChat free, 100 conversations, no card.

Found this useful? Pass it on.
Someone’s funnel is leaking right now.
𝕏 in
Tom Whitfield
Web & Measurement Lead
Tom Whitfield

Tom leads web builds and measurement at Gyrodile: redesigns that protect existing traffic, and dashboards that agree with the bank account.

More insights from the Gyrodile team

More from the playbook

AI Automation AI Chatbot Integration for Websites: Performance, Privacy and Escalation 9 min read AI Automation Enterprise AI Chatbot Development Service for Websites: Build or Buy? 9 min read AI Automation Building AI workflows for Ad Creatives via Nano Banana and VEO 2 min read
Gyrodile
Website in 48h Website Redesign Google Ads Agency GEO Services Creator Marketing Privacy Policy Terms Security
© 2026 Gyrodile Media OPC PVT LTD