Solutions / AI apps & agents
Put an agent at the center of your app
A tutor that teaches your material, an intake assistant that fills your pipeline, support that reads the customer's own order — grounded in your knowledge, scoped to each signed-in user.
The case for building it
Why build this instead of bolting on a chat widget?
A chat widget floats on top of your app and knows none of it: not who's signed in, not what they bought, not what your refund policy actually says. It answers like a stranger reading your homepage. Kliv's agents are part of the app itself — they answer from the knowledge base you author, they see the signed-in user's own data and nothing else, and they act through the same permission system as every page.
The model is the easy part. The product is everything around it: whose data the agent may read, what it's allowed to do, what it should refuse, how its knowledge stays current. Describe those rules and they become the system — an agent with your voice, your limits, and a database-enforced view of exactly one user's world.
Anatomy of the system
What an AI app actually knows
An AI app is not a text box wired to a model. It's a system with opinions about knowledge, identity, and permission — and those opinions are what make its answers worth paying for. Here is the shape Kliv would build.
The agent
Its instructions, voice, and limits — authored by you, changed like the rest of the app.
The knowledge base
What it answers from: your material, in your words — update it and the agent is current.
Users & their data
Each signed-in user's own records — the context that turns a generic answer into their answer.
Conversations
Transcripts tied to the user who had them — records in your database, reviewable by you.
Permissions
What the agent may read and do — enforced by the same rules as every page in the app.
Actions
The things it can do beyond answering — look up an order, book a slot, file a request — each a capability you granted.
Who sees what
Rules the database enforces
- The agent answers the asking user from that user's rows — never from another customer's.
- Knowledge comes from the base you author — answers stay inside the limits you set.
- Every conversation is a record — you can read what your agent actually said.
- The agent's actions go through the same permission system as a human user — no side door.
An agent like this isn't a feature bolted onto the app. It's a user of your app with special duties — same database, same rules, same walls.
What Kliv assembles
A shopping list of platform pieces — each one built in, each one composing with the rest.
An agent in the app
Talks to visitors and acts for the signed-in user — scoped by the same permission rules as every page.
Knowledge you author
The agent answers from a knowledge base built on your material — update it and the agent is current.
Charge for it
Subscriptions gate the agent, meters count its use — the same catalog as everything else you sell.
Every conversation, kept
Transcripts are records in your own database — read what your agent actually said.
Actions that reach your stack
A conversation can end in a Trello card, a booked slot, or a Slack ping — capabilities you granted.
Ways to charge
Three ways an AI app can charge
An agent people rely on is worth money — and the shapes are the same commerce machinery as everything else Kliv sells.
Subscription
$19/month
The member who wants the agent on call
The agent, the member content it's grounded in, and the history of their conversations.
- Unlocks
- An active subscription opens the agent and the gated knowledge behind it; lapse closes both.
- Billing
- Renews monthly; failed cards get retries and recovery emails before access narrows.
- Refunds
- Approval-gated — a refund is a decision, not an accident.
Metered usage
$0.50 per consult
The user who pays for what they use
A base plan plus a meter on agent sessions — the bill tracks the value.
- Unlocks
- Usage counts against the signed-in user's activity, visible on their billing page before the invoice lands.
- Billing
- Metered consults accrue and bill with the subscription — the meter and the invoice are the same system.
- Refunds
- Disputed line items resolve through the same approval flow.
The free front door
$0
The prospect the agent qualifies
Public answers from your FAQ, plus intake — the agent as your most patient salesperson.
- Unlocks
- A captured record in your own database, with the context of what they asked.
- Billing
- Nothing is charged — the value lands in your pipeline instead.
- Refunds
- Nothing to refund; the only thing collected is the lead.
Before launch
The first week, before you launch
An agent goes live in minutes; an agent you trust in front of customers takes a week of honest rehearsal.
Shape the app
Describe the product and the agent's job: what it knows, whom it serves, what it may do. Kliv builds the first preview with the agent already answering.
Give it your voice
Author the knowledge base from your material, set the tone and the limits — what it answers, what it declines, when it hands off to you.
Load the real thing
Real content, the data the agent acts on, the offers if you're charging. Invite a colleague whose job is to ask unfair questions.
Try to break it
Ask it what it shouldn't know. Sign in as two different users and confirm each gets their own world. Then read the transcripts — all of them.
Then rehearse a year
If the agent is a paid product, rehearse the business in the Sandbox with simulated money and simulated time: subscriptions renew, a card fails in March and recovers in April, a cancellation closes the agent on the right date. The launch is a formality because the year already ran.
Bespoke, without the bespoke risk
Custom software used to mean a budget, a contractor, and a maintenance contract. This is a different deal.
Enterprise-grade parts
Sign-in, permissions enforced at the database, billing, audit, email — the components under your bespoke system are the same hardened pieces every Kliv app runs on. You compose proven infrastructure; you don't commission code from scratch.
Changes ship in order
When your process changes, say so — the change is built in conversation, you review it in a live preview, and nothing reaches your users until you publish. Commerce changes can even be rehearsed in the Sandbox with simulated money and time.
Nobody can sunset it
Off-the-shelf tools get re-priced, redesigned, and retired on a vendor's schedule — and your workflow goes with them. Your Kliv app changes only when you change it. It's your product, not a seat in someone else's.
Say it in one sentence
“Build a cooking site where a chef assistant answers from my recipes, remembers each member's dietary preferences, and suggests this week's menu from what's in season.”
This is genuinely how it starts — describe it, then refine in conversation.
Build or buy
Why not just bolt on a chat widget?
| Off the shelf | With Kliv | |
|---|---|---|
| What it knows | Your homepage and whatever docs you pasted in. | The knowledge base you author, plus the signed-in user's own records. |
| What it can do | Deflect tickets. | Act: look up the order, book the slot, file the request — through capabilities you granted. |
| Whose rules | The widget vendor's guardrails and tone. | Your voice, your limits — changed like the rest of your app. |
| The data trail | Conversations in a vendor's dashboard. | Transcripts as records in your own database. |
| The seam | A floating iframe over your app. | A participant inside it, behind the same permissions as every page. |
What it knows
Off the shelf: Your homepage and whatever docs you pasted in.
With Kliv: The knowledge base you author, plus the signed-in user's own records.
What it can do
Off the shelf: Deflect tickets.
With Kliv: Act: look up the order, book the slot, file the request — through capabilities you granted.
Whose rules
Off the shelf: The widget vendor's guardrails and tone.
With Kliv: Your voice, your limits — changed like the rest of your app.
The data trail
Off the shelf: Conversations in a vendor's dashboard.
With Kliv: Transcripts as records in your own database.
The seam
Off the shelf: A floating iframe over your app.
With Kliv: A participant inside it, behind the same permissions as every page.
Common questions
What does the agent actually know?
Two things: the knowledge base you author — your material, updated by you, current the moment you save — and the asking user's own data. Not the open internet's opinion of your business.
Can it see one user's data and not another's?
Yes — that's the point. The agent reads through the same row-level rules as every page, scoped to the user who's asking. Member A's dietary preferences never season member B's menu.
Can the agent do things, not just answer?
Yes — through capabilities you grant explicitly: look up the customer's own order, create a request, book a slot, post to the tools you've connected. No grant, no action.
Can I read what it said?
Every conversation is a transcript in your own database. Review them, learn what people actually ask, and feed the gaps back into the knowledge base.
How do I keep it on-message?
Its instructions, tone, and limits are yours to author — what it answers, what it declines, when it hands off to a human. It speaks in your voice because you wrote the voice.
Can I charge for it?
Yes — the agent is gated like any feature: a subscription opens it, a meter counts its use, and both land on the same billing page as everything else you sell.
