AI assistant platform
Kav identifies customers, pulls live data from your CRM, billing and knowledge systems, and answers on your website, Telegram and WhatsApp — accurately, securely, and handed to a person the moment that's what's needed.
No account needed — this is Kav's own assistant, answering from what Kav publishes.
Four guarantees a prompt bolted onto your website can't make.
Every amount, date and status a customer sees is rendered from your own data through a typed template — the model chooses which template to show and writes the words around it, but it never authors the figure itself.
A capability is a configured connection to your CRM, billing, scheduling or knowledge systems. Adding one is a database change, not a new deployment — nothing here is hand-built per organization.
The same capability answers on your website, Telegram and WhatsApp without being rebuilt. Add a channel and every existing capability is already there.
Personal data is classified per field and gated by one disclosure boundary, and every staff view of it is an audit record with a named actor — the same discipline a customer-facing system needs no matter who's asking.
Eleven questions. Eleven completely different kinds of organization. One mechanism underneath, unchanged. The first is a live municipal deployment; the rest show exactly what the same engine looks like anywhere else — a utility, a clinic, a store, a university, a service firm, a phone carrier, a retail chain, a law firm, an insurer — installed with new database rows, never a line of new code.
On your website, in Hebrew, Arabic, Russian, Amharic or English. A question about personal data opens a sign-in on the server — the amount comes straight from your billing system, never from the model.
A grid a screen reader can navigate row by row and column by column, because this platform's own accessibility rules are explicit that a data grid is a real table element or it isn't accessible. The capability declares its output is tabular; the widget draws the columns the capability's author named — it never guesses a layout from the numbers.
The capability said only "one card per thing" — CardList, the same render hint the tax balance card used in step one. Past four results the widget decides on its own to make it a swipeable rail rather than a long scroll, and does it with no auto-advance and no timer, so nothing here can fail the same accessibility rule a real carousel usually breaks.
Past eight options the same form field stops being a dropdown and becomes a search box — the identical mechanism a thousand-street address register uses, seen for the first time on a live municipal deployment. Locked options are shown, not hidden, with the reason why; a resident who types something the list doesn't contain still gets to send it as an ordinary message, never forced into an answer that isn't theirs.
Every label, icon and destination on this sheet is a row an operator wrote in the Studio's own menu editor — the exact same mechanism the "conversation menu, as a diagram" behind-the-scenes view shows from the staff side. A blank composer is a bad front door; this teaches the words that actually retrieve, and a locked item is shown rather than hidden, so nobody is surprised by the sign-in it will ask for.
A completely different organization, the exact same mechanism: a billing capability bound to the utility's own system, an inline keyboard instead of website buttons. The model chose which numbers belonged on screen — it never wrote a single one of them itself.
Health-adjacent modules are floored to the strictest sensitivity class before a single record is ever read — never sampled into analytics, no summary written if the conversation reaches a person. Illustrated here; a clinic runs it on its own WhatsApp Business account, once Meta has approved that account.
Identity federates to your own login; Kav never holds a password and never takes an account number out of the chat. The order card comes from your order system through the exact same typed contract as the tax balance in step one.
Eligibility is a named question answered by a rule table your registrar edits directly — compiled and tested before it's ever published. The answer that comes back is a decision with its own reason attached; the model didn't decide it, and it didn't write it.
A write never runs on the model's say-so, anywhere in this platform. Kav shows exactly what it's about to do and carries out only what you confirmed — never a re-guessed version of it, and never twice.
A scheduled sweep reads each source's own cadence and re-crawls it automatically — a change on the insurer's own policy page becomes a *draft* article, never a published one. A named person in the accountable department has to submit it and a *different* person has to approve it before the assistant can ever read it back to a resident; the same two-person rule that guards the tax rules in step one guards this text too.
The same platform, from the staff side: a no-code conversation diagram, one queue for every channel, the same conversation inside your CRM, a content-approval queue, a live flow simulator, editable decision tables, contrast-checked branding, and model tiers your team controls.
Every menu, question and handoff is a row in the Studio's own database — not code. A published version is frozen; a change opens as a new draft.
Conversations from the website, Telegram and WhatsApp land in the same queue, by team. An agent doesn't need to know which channel a conversation came from to answer it.
The Kav panel opens inside Microsoft Dynamics 365, so an agent who lives in the CRM all day takes a conversation without leaving the record.
A crawl proposes a draft; it never publishes one. A named person in the department submits it, and a *different* person has to approve it — the same two-person rule reused everywhere else content changes.
Every source — a page, a file, a feed — carries its own re-crawl cadence and its own overdue flag, so a stale page is a thing your team sees, not a thing a resident discovers first.
A live simulator walks the exact published flow as five different residents — a verified owner, an unregistered visitor, a blocked account — with the real validators and the real copy, so "what happens if…" is something your team watches rather than guesses.
Eligibility, discounts, routing — each is a named question answered by rows your team writes, not a paragraph of code. A near-miss explains itself: "age was 63, needed 65," not just "no."
A login page failing accessibility contrast isn't a style preference — it's a legal obligation for a public body. Every colour pair here is checked at save time, with a live preview beside it, before a city can publish.
A cheap model for a simple lookup, a stronger one when the question needs it — set once per city, with a spend ceiling your team can see filling up, never a hidden default.
Capabilities, connectors, a rule engine and approved knowledge are one generic mechanism: any organization with its own systems and its own knowledge base installs it with new database rows, not new code. Municipal government — resident services, property tax, permits — is Kav's first proven deployment; a company support desk, a healthcare provider or a university runs on the exact same engine.
One typed action — "what's my balance", "book a slot" — defined once and executed against your systems through a connector: REST, SOAP, SQL, GraphQL or a webhook.
Hybrid search plus a confidence threshold: below it, Kav says it doesn't know and offers a person, rather than guessing.
A decision is a table your team edits, compiled to run in memory — never a runtime guess and never a code deploy.
Website, Telegram, WhatsApp — one capability answers on all of them, so a new channel is configuration, not a rebuild.
The same engine with different systems behind it. Each row opens the case for that industry: what a customer asks, what Kav answers from your own system, and a worksheet where you supply the numbers.
| Industry | Fewer repeat contacts | Faster when a person steps in | Out of hours, every channel | Fewer errors and rework | Revenue kept or captured | Compliance and audit |
|---|---|---|---|---|---|---|
| Utilities and energy | Main lever | Also covered | Main lever | Main lever | Also covered | Not covered |
| Healthcare providers | Main lever | Also covered | Main lever | Also covered | Also covered | Main lever |
| Retail and e-commerce | Main lever | Also covered | Main lever | Also covered | Main lever | Not covered |
| Higher education | Main lever | Also covered | Main lever | Main lever | Also covered | Also covered |
| Professional services | Also covered | Main lever | Main lever | Main lever | Main lever | Also covered |
| Telecom and internet providers | Main lever | Also covered | Main lever | Also covered | Main lever | Not covered |
| Insurance | Main lever | Main lever | Main lever | Main lever | Not covered | Main lever |
| Banking and financial services | Also covered | Also covered | Main lever | Main lever | Also covered | Main lever |
| Property and housing management | Main lever | Also covered | Main lever | Also covered | Also covered | Not covered |
| Logistics and delivery | Main lever | Also covered | Main lever | Main lever | Also covered | Not covered |
Main lever in that industry's article Also covered
This is a reading guide, not a savings claim. No customer is named and no percentage appears here — every article ends with a worksheet where you enter your own volumes and costs.
What works today, and what's on the way.
Articles on what Kav can do, and a full Knowledge Base on configuring it — organized the way you're actually looking for them.
AI shopping assistant for online shops: product cards, order status. Available for WooCommerce, PrestaShop, Magento, Shopify, BigCommerce and Wix Stores.
How Kav is being built to answer deal, case and invoice questions from your CRM or ERP through common contracts. The contracts exist; vendor connectors are planned.
A 13-minute video course with a reference table: every question in the Services and treatments template, from the department to the install.
What Kav is, who it is built for, and how it differs from helpdesk suites, DIY chatbot builders and WhatsApp inbox tools, with the cases where it is not the right fit.
A 15-minute video course for new agents, with 13 chapters and a full transcript: status, the queue, the copilot, privacy, replying, closing and callbacks.
No. The model never writes a number. Amounts, dates, reference numbers and statuses reach the customer through a display template, straight from your own system's output. The model picks the template and writes the sentences around it.
When a question touches personal data, the customer signs in — with a one-time code or through your own identity system. Identity is decided on the server from the verified session, never from a number typed into the chat. Every capability declares the level of assurance it needs.
Kav keeps no copy of your databases: every personal question is a live call to your own system. Every field has a sensitivity class — financial, identity, legal or minors, for example — that decides what is kept, what enters analytics and what an agent sees. Every staff view of personal data is recorded, with who looked and why.
No self-service signup — a platform operator sets up every organization directly, in three steps.
A person reads every request — talk to us, and we'll take it from there.
Try the assistant now — no account needed. It's Kav's own assistant, answering from what Kav publishes.
