Skip to content
Huematte

Support agents that already know your domain

“I want to change this.”means three different things.

An exchange in e-commerce. A prorated downgrade in SaaS. A new slot at a clinic. Huematte agents are pre-trained on each domain's patterns, then grounded in your policies, so answers fit your business instead of a generic guess.

See an agent in your domainContact sales
Returns & Orders
P

Priya S.

Web chat

Returns & Orders agent
I want to change this.
Customer, just now
Understood as: Exchange request
Order #4821Delivered 9 days agoWithin 30-day window

Returns policy, section 2.1. Unworn items can be exchanged within 30 days of delivery.

I can start an exchange for you. Which size would you like?
Huematte agent, cited 1 policy

Exchange drafted

Linen overshirt, size M to size L. Pickup booked on confirmation.

Size SSize LSize XL
Agent is handling this conversation. A teammate can take over any time.
Pre-trained
on domain patterns, not started from a blank prompt
Grounded
in your own policies, with the source on every answer
Handed off
cleanly, with full context, when a person should decide

The problem

A polite non-answer is still a non-answer.

General-purpose chatbots are fluent, but fluency isn't knowing that a return window, a proration rule and a reschedule policy are three different problems.

One brain for every question

A generic assistant treats “change this” the same way everywhere, so it asks clarifying questions an experienced agent in your industry would never need to ask.

Months of tuning before it's useful

Teaching a general model your domain's edge cases means writing prompts, scripts and exceptions one ticket at a time, and repeating it every time a policy changes.

Then a ticket, then a wait

When the bot can't act, the customer gets a vague reply, a ticket number and a queue. The person who picks it up starts from zero.

The domain lens

Same words. Different jobs.

Customers don't describe their problem in your team's vocabulary. Each Huematte agent reads the same short message through what its domain usually means by it.

“I want to change this.”

E-commerce

Exchange for another size, inside the 30-day window

SaaS

Downgrade to Starter with a prorated ₹1,240 credit

Healthcare

Move the Thursday slot, within the 12-hour rule

“Where is it?”

E-commerce

Live courier status and the promised delivery date

SaaS

The latest GST invoice, sent to the billing contact

Healthcare

Clinic location and room for the booked visit

“I was charged twice.”

E-commerce

Duplicate UPI payment matched to one order, refund started

SaaS

Two card charges found, handed to billing with IDs

Healthcare

Booking deposit checked, passed to the front desk

“Cancel it.”

E-commerce

Cancel free before dispatch, or book a return after

SaaS

Cancel at term end, with access kept until then

Healthcare

Cancel the visit and offer the slot to the waitlist

Compare the three agents side by side

How it works

From a pre-trained agent to live, in four stages.

You don't build a bot. You pick the agent for your domain, point it at your rules, and check its work on your own history.

The full process
  1. 01

    Choose a domain

    Start from an agent that already knows your vertical's intents, vocabulary and edge cases, not a blank prompt.

    Returns & Orders
    Billing & Accounts
    Scheduling
  2. 02

    Ground it

    Connect your real return windows, plan rules, catalog and clinic calendars. The agent answers from these, and only these.

    returns-policy.pdfSynced
    shipping-rules.mdSynced
    Order data feedSynced
  3. 03

    Test on past tickets

    Replay your own resolved conversations. Compare each agent answer with what your team actually said before anything goes live.

    Ticket #18820 Matches team answer
    Ticket #18807 Matches team answer
    Ticket #18794 Would hand off
  4. 04

    Go live

    Turn it on for web chat and email. It resolves what it can, and hands off what it shouldn't, with the full context attached.

    Agent live

    Web chat, Email

    Handoff: over limit, unclear, or asked for a person

What the agent actually does

Not a script. Not a guess.

Four things every Huematte agent does on every conversation, in this order.

Domain-pretrained understanding

Knows the common intents, edge cases and vocabulary of its vertical from day one. “Size swap”, “wrong fit” and “change this” all land on the same exchange flow.

“size swap please” Exchange
“wrong fit, need bigger” Exchange
“I want to change this.” Exchange

Policy grounding

Answers cite your own return, billing or scheduling rules. If no rule covers the question, the agent says so and hands off instead of inventing one.

Your ₹1,240 credit will be applied to the 25 Oct invoice.

billing-policy.pdf, section 4.3Mid-cycle plan changes are prorated to the day as account credit.

Action within limits

Starts an exchange, applies a credit or moves a slot, but only inside the boundaries you set. Anything over a limit is prepared, then passed to a person to approve.

Refund without approvalup to ₹5,000
Account creditup to ₹2,000
Reschedules per bookingup to 2

Clean human handoff

When a case needs a person, your teammate receives the conversation, the detected intent, the records checked and the reason for handoff. Nobody asks the customer to repeat themselves.

Handoff note to Billing team

  • Intent: duplicate charge, card refund
  • Checked: two charges of ₹4,800 on 25 Sep
  • Reason: card refunds need approval

The domain agents

Three verticals, each with its own hue.

Every agent ships with the intents and edge cases of its domain already mapped. Pick the one that matches the questions your team answers most.

E-commerce Returns & Orders

Returns, exchanges, order status, refund eligibility and address edits, checked against your own windows and conditions.

  • Return-window checks
  • Size and colour exchanges
  • Courier tracking
  • Refund eligibility

SaaS Billing & Accounts

Plan changes, proration, invoices and failed payments, answered with the exact figures from your billing rules.

  • Upgrades and downgrades
  • Proration and credits
  • GST invoices
  • Failed-payment recovery

Healthcare Scheduling

Booking, rescheduling, reminders and waitlists. Scheduling and admin only. Clinical questions go straight to your staff.

  • New bookings
  • Reschedules and cancellations
  • Reminder settings
  • Waitlist offers

Custom Domain Agents

For verticals not covered yet: insurance claims, travel changes, education admissions and more.

How a new agent is built

Scope boundaries

Knowing what not to answer is part of knowing the domain.

Each agent has hard edges that you can tighten but never loosen past. These are written into the agent, not left to a prompt.

Security and data handling

Returns & Orders

Never

  • Refund above your approval limit
  • Override a final-sale rule
  • Change an order after dispatch

Billing & Accounts

Never

  • Refund to a card without approval
  • Waive an invoice
  • Invent a discount

Scheduling

Never

  • Give medical advice
  • Interpret symptoms or reports
  • Book outside clinician rules

Support analytics

See why conversations end the way they do.

Resolution by intent, the exact reasons for each handoff, and which questions are growing week to week, split by domain.

  • Every handoff tagged with its reason
  • Every action logged against the limit that allowed it
  • Gaps in your policies surfaced as they appear
How support analytics works

Handoff reasons

Sample view

Over action limit

No matching policy

Outside scope

Customer asked for a person

Illustrative layout. Your dashboard shows your own conversations.

Try the lens

Pick a domain.
Type a customer question.
See it understood.

A demonstration on sample data. Connect your own policies to see your real answers.

Pick a domain

Returns & Orders agent reads this as: Exchange request

Order #4821Delivered 9 days agoWithin 30-day window
I can start an exchange for you. Which size would you like instead?

Returns policy 2.1: exchanges within 30 days of delivery

Outcome

Resolved within limits

Exchange drafted, pickup booked on confirmation

Run it on your policies