Set up booking: every option in the wizard, a video course for administrators
A 13-minute video course with a reference table: every question in the Services and treatments template, from the department to the install.
Knowledge Base
Everything you need to configure Kav Studio yourself, end to end — written for the person doing the configuring, not the person who built the engine.
Getting started
Orientation: how the Studio is organized, and what publishing a draft actually does.
How Kav compares with other chatbots and helpdesk tools
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.
What are solution templates? A video course for tenant administrators
A 6-minute video course with a transcript: what a solution template is, the gallery, the wizard, the review, the install and what to do after.
Finding your way around Kav Studio
A tour of the Studio's modules, and the one distinction that matters everywhere in it: a draft you're still shaping versus what residents actually see.
Connecting your first system, end to end
The full path from a bare API to a working assistant answer: a connector instance, a declared capability, and the gate that checks both before anything goes live.
Capabilities
Wiring one action — a balance, a booking, a lookup — to your own system, and choosing how its answer looks.
Choosing how an answer looks: render hints
Text, CardList, Table, Detail or Confirmation — the render hint a capability declares decides its shape everywhere it's used, and the model never draws it from scratch.
The three faces of a capability, from the authoring screen
Semantics, contract and policy are three separate decisions on one form — here's what each one actually controls, and why the contract is frozen the moment it publishes.
Previewing a capability exactly the way the model sees it
The tool tab compiles the real, per-city tool definition live — the same fingerprint, the same stripped identity fields, the same JSON the model is actually handed.
The AI module, explained field by field — for absolute beginners
What use_when, do_not_use_when and every other field on a capability's AI-facing form actually does, with no technical background assumed.
Catalogue management
Module versions, language coverage, licence grants, the fact catalogue, and the audit ledger behind every capability.
Why a capability answers in Hebrew and not English, and how to fix it
A capability can publish complete in every other way and still be unreachable in one language — the coverage report finds the gap and fills it without deleting anything.
Granting a licence bundle — and why it doesn't switch anything on by itself
A commercial licence grant gates whether a module pack is even allowed to install, but granting one to an already-installed, disabled module changes nothing on its own.
Module versions: draft and published are not two states of a switch
A module version is where a capability gets somewhere to live — draft is fully editable, published is frozen, and there's no un-publish.
Reading the audit log, and why retention can only ever go up
Every staff view of a resident's personal data is already recorded — this screen is how anyone reads that ledger back, and reading it is itself an audited act.
The platform fact catalogue: what a rule condition is allowed to read
A fact like resident.age means the same thing in every rule — registering one is a platform decision, and checking references first stops you orphaning a live rule.
Connectors
A connector's own lifecycle — draft, mock and live — and how a credential is referenced without ever being shown again.
Connect your own tools in the Studio, step by step (no code)
A plain walk from your first connector to a live, tested answer: connector, capability, rules, assistant settings, test and publish. Real Studio screens.
A connector's lifecycle: why every instance starts Draft and Mock
Going live is a separate, deliberate act from creating a connector — every new instance seeds unable to call anything, by construction, until someone says otherwise.
Referencing a credential you'll never see again
The connector screen that touches real secrets shows a vault key name, a fingerprint and a last-used time — and deliberately nothing that could ever be the value itself.
Channels
Turning on the website widget, Telegram and WhatsApp, and keeping every capability answering on all of them.
Connecting Telegram, WhatsApp and the rest
Registering a channel surface makes it visible, not live — and WhatsApp, Messenger and Instagram share one Meta adapter with one reply-window rule between them.
Turning on the website widget
One script tag, a widget key that decides everything server-side, and a branding screen with contrast checks that block publishing a page nobody could read.
City settings
Sign-in methods, model tiers and policy, hand-off phone numbers, and every organization-wide setting that isn't a capability.
Registering a model credential and choosing your tiers
How to register a model provider, activate it, choose your cheap and strong tiers, and set a spend ceiling per billing period.
Enabling a sign-in method for your residents
How to add an authentication provider in the Studio, choose its trust model, and understand why one switch on this screen needs its own permission.
Overriding the widget's platform wording for your organization
How to see and edit your own per-key text overrides for the widget's built-in strings, without touching what other cities see.
Setting your answer policy and retrieval grounding
Configuring how your assistant answers knowledge questions and how strictly it grounds them, with every row showing its effective value and source.
The phone numbers your assistant can hand off to
Configuring your call centre and crisis-support contact blocks, in every language you publish, and why the server refuses a digit in the wrong field.
The widget's quick-menu editor
How to build the tappable shortcuts a resident sees before typing anything, with a live preview that drills the same way the real widget does.
Turning optional model steps on and off
How to activate a platform model task for your city, override its wording, and read the effective-state pill that tells you what's actually running.
Widget identity, delivery, languages and greeting copy
Configuring the widget key, loader origin, entry flow, bundle key, published languages and per-language greeting text in one screen.
Widget placement, and where the colors actually come from
How to set the launcher's corner and open-on-load behavior, and why this screen deliberately can't edit your palette.
Knowledge
Adding sources, setting a re-crawl schedule, and the two-person review every article passes through.
Adding a knowledge source and its schedule
Four questions decide a source's whole record: which part of your site this is, what it's about, how often to re-check it, and how far the crawl may go.
Answer overrides: when you need the exact words, every time
An override doesn't rank higher than retrieval — retrieval doesn't run at all. That's why it's guarded more tightly than an ordinary article.
Departments and approvers: why content can get stuck with nobody to approve it
A department with zero approvers isn't a warning — every article it owns is permanently unapprovable until someone is appointed.
Previewing retrieval: what the assistant would say, and why
Type a real question and see the exact passages, rankings and overrides your assistant would use — before a single resident ever asks it.
Reading your knowledge health dashboard
One screen answers the question every other one leaves scattered: what's broken, what do we have, and is any of it stale.
Testing one article the way the widget would render it
The Ask tab runs a real assistant turn and draws the answer with the widget's own renderer, inside the same isolated shadow root residents actually see.
The two-person approval queue, and why three gates guard it
A crawl proposes a draft; it never publishes one. Submitting and approving an article are always two different people, and the queue enforces that three separate ways.
Which languages you publish content in
Turning on a resident-facing language isn't a free toggle — it's a recurring translation bill, and the screen shows you the number before you commit.
Localization
Finding untranslated copy, the glossary that feeds both rendering and search, and sending text to a translation vendor.
Checking each language's translation coverage at a glance
One table lists every language your platform could offer, including the disabled ones, with its own translation completeness beside it.
Finding untranslated copy before it ships
An automated check fails the build over any interface text missing from an enabled language — this report is the list that stands between you and a green build.
Sending copy to a translation vendor, and bringing it back
Export a CSV, hand it to a bureau or a bilingual staff member, and re-import it as a preview you approve — never a direct write to your live catalogue.
The glossary: one table, two jobs
One glossary entry fixes how a term reads in your copy and expands what a resident's question can match in search — confusing the two is the mistake to avoid.
Rules engine
Building and testing a decision table — eligibility, discounts, routing — without writing code.
Reading the conflict-analysis panel
Three read-only views of one decision table, a live shadowed-rule warning, uncovered-input gaps, and the stats that change which rule you'd write.
Testing a rule before you publish it
Publish is blocked until every gate clears — analysis errors, failing tests, an uncovered input with no default, and an impact preview you have to actually run.
Building your first eligibility rule
A decision table your team edits directly — four views of one document, and analysis that runs continuously rather than on a button you have to remember to press.
Conversation flows
Designing menus, forms and handoffs as a diagram your team edits directly.
Building a form step, with real validation
A form node splits into three tabs — structure, preview, and timing — and publish is blocked on the accessibility half you had the least reason to open yourself.
Building multi-level menus with submenus
A menu option's target is any node in the flow, including another menu — so a submenu isn't a separate feature, just where you point one option.
Creating a form, field by field
Twelve real field kinds, five vetted patterns, and options that come from a capability instead of a list you maintain by hand — never a free-text validator.
Designing a conversation menu with no code
A menu node is a question and its branches, edited as a real diagram, a real grid, or a document — four views of the exact same flow, never four copies of it.
Named forms vs. a flow's own form step
A form node lives inside one flow. A named, standalone form is its own object the assistant can find directly from what a resident actually types.
Testing a flow with the simulator
Five personas, real validators, and copy resolved from the exact catalogue the translation gate checks — walking a flow before a resident ever does.
Writing retry and quota messages
Every attempt quota needs an extension and a real exit — the editor makes both the default, not an option an author has to remember.
Security & access
Assurance levels, entitlements, and who on your staff can see what.
Setting assurance levels and entitlements on a capability
Every capability declares a floor: how sure the platform must be of who's asking, how sensitive the data is, and any extra entitlement check beyond identity alone.
Staff roles and permissions: who can see what
A role decides what a token can reach across the whole platform. Granting or revoking one is a single deliberate act — there's no draft, and no batching.
Desk & analytics
Configuring the agent desk's queues, and the model-tier and spend controls your team can see.
How to use the Agent Desk: a video course for new agents
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.
Configuring the agent desk's queues
A queue's whole policy is five fields: which department owns it, how the bot behaves during a handoff, a sensitivity ceiling, and whether it's actually open.
Model tiers and spend ceilings
Registering a model credential and putting it into service are two separate acts, and there is no field anywhere in the Studio that could hold a raw API key.
Staffing your desk: agent clearances
A permission grants a verb. A clearance grants a scope. Both are required independently, and your own row is the one row you can never change.
Wrap-up codes: tagging how a conversation ended
Five scalar fields make a wrap-up code, one immutable — and the authoring screen deliberately isn't scoped to your own department.
Kav Desk for Dynamics
Onboarding the Dynamics 365 side panel, and maintaining a live surface's origins, scopes and erasure ledger.
Maintaining a live Dynamics surface: origins, scopes, and erasure
Day-two operations for an already-onboarded Dynamics panel: managing origins, linking staff identities, scoping access, and the erasure ledger.
Onboarding Kav Desk for Dynamics: the 13-step wizard
The only wizard in the whole Studio — 13 steps across four services, from Entra directory binding to a published side panel inside Dynamics 365.
Campaigns & outbound
Creating an outbound campaign, its message templates, audience import, consent, suppression and quiet hours.
Reading the consent and suppression ledgers
Two different stops on an outbound message, shown as two separate tables — one records what a resident agreed to, the other overrides it regardless.
Creating your first outbound campaign
A campaign is one form — its purpose, its audience segment, and a channel ladder that decides where a message goes first and what happens if it isn't opened.
Importing an audience from a CSV
Upload a resident list into a Manual segment: map exactly one column to a resident key, an optional language column, and everything else becomes a merge field.
Setting quiet hours for outbound messages
Windows when your organization won't contact residents, in a local time zone — and why an unscoped policy behaves very differently from a channel-scoped one.
Writing a campaign message template
One composer for every channel's message body, merge fields that stay honest in preview, an optional AI draft pass, and two lifecycles that move independently.