Put an AI agent in your app
Add an assistant that talks to your visitors and acts on their behalf — answers questions, looks things up, updates their records. Describe what it should do and what it's allowed to see; Kliv runs it.
Agents that act
Not a bolt-on bot — part of your app
Your agent lives inside your app, working with the same data and the same logic you've already built. It can do real work for your visitors, safely scoped to each one, with no model to host and nothing to wire up. You stay in control of what it does and who can use it.
The anatomy of an answer
What separates an agent from a widget is everything that happens between the question and the reply. Here is that path, step by step.
A question arrives
A visitor asks in plain language — "where's my order?", "what does the Pro plan include?" — in a conversation that is part of your app, not an iframe floating over it.
Identity first
Before anything else: who is asking? The agent knows whether this is an anonymous visitor or a signed-in user — and exactly which one.
Permissions applied
The agent reads through the same access rules as every page in your app. It can see this user's rows — their orders, their bookings — and nobody else's. The boundary is enforced underneath the agent, not written into its prompt.
Grounded, then answered
It answers from the knowledge base you author — semantically indexed, so it finds the right article by meaning, with no vector database for you to run — plus the asking user's own records. Your voice, your limits: what it declines is as designed as what it says.
Or it acts
Some answers are actions: create the entry, update the profile, file the request. Actions run through your app's own server-side functions — capabilities you granted explicitly, executed as the asking user, nothing implied.
The transcript lands
Every conversation becomes a record in your own database. Read what your agent actually said, see what people actually ask, and feed the gaps back into its knowledge.
An assistant that actually helps
Configure it in minutes. Kliv handles the AI.
Talks to your visitors
A conversation built into your app. Your visitors ask in plain language and the agent answers in your app's voice — from instructions you give it and a knowledge base you author.
Acts on their data — safely
The agent can look things up and make changes on the signed-in visitor's behalf, scoped so it only ever touches that visitor's own records — never another customer's data.
You stay in control
Decide who can talk to it, what it's allowed to do, and how much it can run. Access starts closed; you open it deliberately.
Set up in minutes
Give it a name, write its instructions, choose who it serves and what it can do. No engineer knobs — just describe the assistant you want.
Grounded in your knowledge
A knowledge base you author, semantically indexed out of the box — the agent finds the right answer by meaning, and it's current the moment you save. No vector database to stand up.
Every conversation, kept
Transcripts land as records in your own database — review them, learn what people actually ask, and answer "what did the agent tell this customer?" with certainty.
Agents that act on your data
It does things, not just chats
Give your agent the ability to work with your app's data, and it can manage a visitor's records for them — find an order, update a profile, create an entry — always limited to the person it's talking to. Kliv enforces that boundary underneath the agent, so even a cleverly worded visitor can't get it to reach into someone else's data.
Runs on the AI you already have
Agents use your account's AI usage — the same budget you build with. Set per-visitor and per-agent limits so a busy day stays in bounds.
How agents workCommon questions
Where do the agent's answers come from?
Two sources: the knowledge base you author — your material, semantically indexed, current when you save — and the records the asking user is allowed to see. It doesn't answer from the open internet's opinion of your business.
How is "only their own data" actually enforced?
Below the agent, not inside its prompt. Every read and write passes through the same permission rules as the rest of your app, scoped to the signed-in user — so a cleverly worded request can't talk it into someone else's records.
Can it do things, not just answer?
Yes — through capabilities you grant: look up an order, update a profile, create a record, call one of your server-side functions. No grant, no action — and every action runs as the asking user.
Can I see what it said?
Yes — transcripts land as records in your database, tied to the user who had the conversation. They're how you audit tone, catch gaps in the knowledge base, and answer "what did we tell this customer?"
How do I keep it on-message?
Its instructions, tone, and limits are yours: what it may discuss, what it declines, when it hands off to a human. Because you author both the instructions and the knowledge, staying on-message is design, not luck.
Can I charge for access to it?
Yes — gate the agent behind a plan like any other feature, or meter its use per conversation. The commerce engine handles both, and the AI-apps deep dive shows the business patterns. Commerce → AI apps →
