Live on your website and Instagram · WhatsApp and email next

Your AI employee for customer conversations

Create an AI agent that understands your business, talks to your customers, and actually gets things done.

Train it on your website and business knowledge. Connect your tools. Let it answer questions, qualify leads, and perform tasks automatically.

No coding required. Free to start, no card needed.

Your AI Agent

Website

Customer

Can I return my order?

AI Agent

Yes. Your order #1024 is eligible for return until 22 August.

Would you like me to start the return?

Start Return
KnowledgeToolsRules

AI activity

  1. Understood customer intent
  2. Checked return policy
  3. Retrieved order #1024
  4. Verified eligibility
  5. Waiting for confirmation

It read the policy, fetched the order and checked the date — then stopped, because starting a return changes something.

The difference

Your customers don’t want another chatbot.

They want answers. They want help. And when something needs doing, they don’t want to be sent to another form, email, or support page.

Traditional chatbot

Answers, then gives up

  1. Customer asks
  2. Searches an FAQ
  3. Generic answer
  4. “Contact support”

Your AI agent

Understands, then finishes the job

  1. Customer asks
  2. Understands intent
  3. Checks your business knowledge
  4. Uses your connected tools
  5. Takes action
  6. Request resolved

Same question. The difference is what happens between reading it and replying — and whether anything actually changes as a result.

What it does

From answering questions to getting things done.

Four capabilities that build on each other. Most tools stop at the first one.

01

Understand

Train your agent on your website, documents, FAQs, products, services and business knowledge. It answers from those and cites them.

02

Recommend

Help customers find the right product, service, plan or course — matched to what they actually asked for, not to a keyword.

03

Act

Connect your business tools and let your agent perform real tasks. Each action carries its own approval rule.

04

Automate

Turn conversations into workflows that keep working after the chat ends — follow-ups, alerts, handovers, scheduled checks.

Getting started

Build your AI agent in minutes.

Four steps, in order. Most businesses have something answering real questions the same afternoon.

01

Add your knowledge

Point it at what you already have. Indexed and answering within minutes.

  • Your website
  • PDFs and documents
  • FAQs
  • Individual URLs
  • Plain text
02

Configure your agent

Plain settings screens. No prompt writing, though you can add your own instructions.

  • Personality and tone
  • Custom instructions
  • What to do when unsure
  • Lead capture rules
  • Escalation triggers
03

Connect your tools

Any HTTP API your business already runs. Each connection gets its own approval rule.

  • Custom REST APIs
  • Order and account lookups
  • Availability checks
  • Ticket creation
  • Named app integrationsOn the roadmap
04

Deploy everywhere

Start on your site with one line of code, then add channels as you need them.

  • Website widget
  • Instagram DMs
  • WhatsAppComing soon
  • EmailComing soon
  • REST APIOn the roadmap

Under the hood

Your AI agent has three parts.

Knowledge tells it what is true. Tools let it act. Rules decide how far it may go. The engine only sits in the middle.

Knowledge

The agent knows what your business knows.

Indexed by meaning, so “do you deliver to Spain” finds a shipping page that never uses the word deliver.

  • Website
  • Documents
  • FAQs
  • Products

Tools

The agent can reach your systems.

Any HTTP API. It sends only the fields you defined and reads back only the ones you asked for.

  • Orders
  • Accounts
  • Availability
  • Custom APIs

Rules

You decide what it is allowed to do.

Per-action approval settings, plus role permissions checked on the server for every request.

  • Permissions
  • Approvals
  • Guardrails
  • Audit log

AI engine

Reads the question, retrieves what is relevant, decides whether a tool would help — and stops where the rules say to stop.

Your customer

A worked example

See what an AI agent can actually do.

A real support exchange, with the reasoning shown. Note the second answer — it does not promise something it has not checked.

Illustration of a typical conversation. Not a live widget.

Customer

My order hasn’t arrived yet.

AI Agent

I can check that for you.

AI activity

Searching order #4821

Order found

Shipping status retrieved

AI Agent

Your order shipped yesterday and is expected tomorrow. Tracking number TRK-928374.

Customer

Can I change the delivery address?

AI Agent

I can look into that. Let me check whether it has already been dispatched — once it has, the courier controls the address.

AI activity

Checking dispatch status

Already with courier — address change unavailable

Offering to raise it with the team

What made that possible

It read your policy

The shipping and returns rules came from your own documents, not from a general model’s assumptions.

It called your API twice

Once for the order, once for dispatch status. Both read-only, so neither needed anyone’s approval.

It refused to over-promise

Changing the address was no longer possible, so it said so and offered the next best thing instead of guessing.

It left a record

Every lookup is logged with what was sent, what came back and how long it took.

Human handoff

When it cannot answer, a person takes over — in the same thread.

No ticket number, no “please email support”, no starting again. The customer keeps talking in the window they were already in, and a colleague answers them there.

The same order as the example above, picked up where the agent gave up.Available

Customer

It says delivered but nothing arrived. Here’s my doorstep.doorstep.jpg

AI Agent

I can’t authorise a replacement myself, so I’ve passed this to our team with the photo and the full thread.

Priya from the team joined this conversation.

Priya

Hi Rahul — I can see the photo. I’ve booked a replacement for tomorrow, no charge. Here is the confirmation.

Priya handed the conversation back to the assistant.

How the handover works

One queue, every channel

Website chat and Instagram DMs land in the same inbox, carrying the whole conversation and whatever the agent already tried.

The agent goes quiet

The moment a colleague takes a conversation, the AI stops replying in it — so the customer never gets two voices answering at once.

Seven days to reply on Instagram

A person gets seven days after the customer’s last message, where the AI gets twenty-four hours. Meta grants the longer window only for human replies, and we use it.

Files both ways

Customers send photos and PDFs; your team sends them back. Every file is identified by its actual contents rather than by what it claims to be.

Handing back is one click, and the agent picks the conversation up again from the next message.

Solutions

One platform. A lot of different jobs.

The agent is the same; the knowledge and the connected tools are what make it yours.

Customer support

Answer the questions that repeat, check an order, start a return, and escalate the rest with the full thread attached.

Sales

Qualify visitors while they are still interested, recommend the right plan, and capture contact details as a scored lead.

E-commerce

Help people find products, answer delivery and returns questions, and look up order status against your own systems.

SaaS

Answer product questions from your docs, walk users through setup, and surface which features people cannot find.

Real estate

Qualify buyers, match enquiries to listings, and capture viewing requests outside office hours.

Education

Answer questions about programmes, fees, admissions and timetables — from your own prospectus, with sources.

Healthcare

General practice information, opening hours and appointment enquiries.

Set up with approval gates and clear boundaries. Not for clinical advice.

Agencies

Build and run agents for several clients, each with its own knowledge, tools and workspace.

For agencies

Run AI agents for every client you look after.

Each client gets their own workspace with their own knowledge, tools and agents. You manage all of them from one login.

A separate workspace per client

Each client's knowledge, agents, tools and conversations are fully isolated. No client can see another's data.

Several agents per client

A support agent and a sales agent for the same business, each with its own knowledge and tone.

Your team in every workspace

Invite your staff into each client's workspace with the access level that fits — including view-only.

Usage visible per client

Every workspace has its own plan, limits and usage screen, so you can see who is growing.

Not built yet

  • White-label branding and your own domain
  • One console listing every client workspace
  • Consolidated billing across clients
  • Reseller pricing and margin controls

If one of these is what decides it for you, tell us — it moves up the list when a real agency needs it.

See agency pricing

How it is structured

Available

Your agency login

Client A

own workspace

Sales agent

Client B

own workspace

Support agent

Client C

own workspace

Lead capture agent

Client D

own workspace

Booking agent

Isolation is enforced in the database, not in the interface — every query is scoped to one workspace, and we test that deliberately.

Integrations

Connect the systems your business already runs.

Today that happens through a connector you point at any HTTP API. Named one-click apps are being built — and until they are, we say so.

Connect anything with an API

Available

Give it a URL, an auth header and a description of what the call does. Your agent can use it from that point on — and you choose whether it runs automatically, asks the customer first, or waits for your team.

Your API
AI agent
Action taken
  • Bearer tokens
  • API keys
  • Basic auth
  • Custom headers
  • Encrypted at rest

Where it can talk to customers

  • Website widgetOne line of codeAvailable
  • Instagram DMsMessenger PlatformAvailable
  • WhatsAppMeta Cloud APIComing soon
  • EmailInbound and outboundComing soon
  • REST APIFor your own appOn the roadmap

Instagram runs against a live business account, replies included. WhatsApp and email are built and still waiting on an account to verify end to end — we would rather label them honestly than have you find out during a trial.

One-click apps on the roadmap

On the roadmap

Shopify

Orders, customers, fulfilment

WooCommerce

Orders and products

HubSpot

Contacts and deals

Salesforce

Leads and cases

Google Calendar

Availability and booking

Zendesk

Tickets and macros

Slack

Internal alerts

Stripe

Payments and refunds

Each of these has a public API, so you can connect it today with the custom connector above. What is coming is the one-click version that skips the setup.

Control

AI that works for you — not without you.

An agent that can act needs limits that are real rather than promised. These are the mechanisms, not a policy statement.

Permissions

Owner, Admin, Member and Viewer. A support agent can work through conversations without being able to rewrite what the agent says.

Approvals

Any action can be set to require a human before it runs. Refunds and cancellations usually are.

Knowledge boundaries

It answers only from material you supplied. No coverage means it says so rather than filling the gap.

Workspace isolation

Every database query is scoped to one workspace. We test that separation deliberately rather than assuming it.

Activity log

Who changed what, who approved which action, and what each tool call sent and received. Written once, never edited.

An approval, as your team sees it

Issue refund

order #4821 · ₹2,000

Needs approval

The agent verified the order, checked it against your refund policy and stopped here. Nothing has been sent to your payment provider.

ApproveReject

Whichever you choose is recorded against your name. “Who approved the refund” has an answer.

Analytics

Know what your agent is doing — and what it keeps missing.

Volume is easy. The useful part is being told which questions your knowledge does not cover yet.

Example figures, shown to illustrate the dashboard layout.
Conversations12,482
Resolved by the agent9,840
Leads captured1,284
Passed to a human312
Resolution rate79%
of conversations finished without a person

Most asked about

Pricing3,120
Delivery times2,310
Refund policy1,624
Product availability1,187

What it could not answer

Questions grouped by what they mean rather than by exact wording — so six ways of asking the same thing show up as one gap worth fixing.

Do you support enterprise SSO?

Asked 17 times, in 4 different wordings. Nothing in your knowledge base covers it.

Can I pause my subscription?

Asked 11 times, in 3 different wordings. Nothing in your knowledge base covers it.

Is there an annual discount?

Asked 8 times, in 2 different wordings. Nothing in your knowledge base covers it.

Type an answer once and it becomes part of the knowledge base immediately. The next customer to ask gets it.

Pricing

Start free. Pay when it is doing real work.

Every plan includes the full agent — knowledge, actions, approvals and analytics. What changes is how much of it you use.

Starter

For a small team running agents in front of customers.

$49/month
  • 5 agents
  • 10 team members
  • 100 knowledge sources
  • 20 connected tools
  • 10,000 messages a month
  • Custom API actions
Start free
Most chosen

Growth

Higher volume, and no cap on how much you build.

$199/month
  • Unlimited agents
  • Unlimited team members
  • Unlimited knowledge sources
  • 100,000 messages a month
  • Multi-agent routing
  • Full audit log
Start free

Enterprise

Negotiated limits. Not shown on the pricing page.

Custom
  • Negotiated limits
  • Multiple workspaces
  • Priority support
  • Security review
  • Onboarding help
Talk to us

A free plan is available to try everything on one agent. Prices and usage limits may change during early access — anyone who has signed up will be told before they do.

Questions

The things people ask before signing up.

Including the ones where the honest answer is “not yet”.

What is the difference between an AI agent and a chatbot?

A chatbot matches a question to an answer. An agent works out what you meant, looks it up in the business's own material, decides whether one of the business's systems could help, calls it if allowed, and then tells you what it did. A chatbot can tell you the returns policy; an agent can check your order against it and start the return.

Do I need to know how to code?

No. Adding knowledge, writing instructions, setting approval rules and putting the widget on your site are all done through the dashboard. Connecting one of your own APIs needs somebody who knows that API — usually a developer for twenty minutes, once.

What can my agent learn from?

Your website, PDFs and documents you upload, individual URLs, plain text you paste in, and FAQ lists you type out. It answers from those and cites which one it used, so you can check it.

What happens when it does not know something?

It says so and offers to pass the question to a person. It is instructed not to fill gaps with guesses, and every unanswered question is recorded so you can see what to write next.

Can my agent actually perform tasks?

Yes, for any system with an HTTP API. You describe what the call does and what it needs, and the agent uses it when a conversation calls for it. Read-only lookups usually run automatically; anything that changes something waits for permission.

Can I control what it is allowed to do?

Every connected action has one of four settings: run automatically, ask the customer to confirm, wait for your team to approve, or switched off. Separately, your staff have roles — a support agent can handle conversations without being able to approve a refund or change what the agent says.

Can I connect my own API?

Yes, and today that is the main way to connect anything. You give it a URL, an authentication header and a description. Credentials are encrypted before they are stored and never shown again.

Is there a Shopify, HubSpot or Salesforce integration?On the roadmap

Not as a one-click app yet. All three have public APIs, so you can connect them today using the custom connector — the packaged versions that skip the setup are being built.

Can customers reach it on Instagram or WhatsApp?

Instagram DMs work today, against your own Instagram business account — the agent answers there, your team can reply there, and photos come through in both directions. WhatsApp is built and has not yet carried a live message, so we are not calling it available; email as a channel is in the same position. We would rather tell you that than have you discover it during a trial.

What happens when the agent cannot answer?Available

It hands the conversation to your team instead of guessing. The whole thread arrives in a shared inbox, and a colleague replies in the same conversation the customer is already in — no ticket number and no starting again. The customer is told when somebody joins and again when the conversation goes back to the agent, and while a person holds it the agent stops replying, so nobody gets two answers at once.

Can customers send photos or documents?Available

Yes, on the website widget and on Instagram, and your team can send files back the same way. Each file is identified by its actual contents rather than by its name or what the upload claimed, and anything not on the allow-list is refused rather than stored.

Can an agency manage several clients?

Yes — each client gets a separate workspace with its own knowledge, agents, tools and usage limits, and your team can be in all of them. White-label branding, a single cross-client console and consolidated billing are not built yet.

How is my data kept separate from other businesses?

Every database query is scoped to one workspace, and that isolation is covered by tests written specifically to try to break it. Connected-system credentials are encrypted at rest, and no certification is claimed — if you need a security review, ask and we will go through it properly.

What does it cost to try?

Nothing. The free plan covers one agent with enough messages to judge whether it works on your own material. No card is needed to start.

Get started

Give your business an AI employee.

Point it at one page of your own material and ask it something awkward. Twenty minutes is enough to tell whether it is any good.

Free to start · No card required · No coding