AI Agents · human in the loop

AI that drafts. People decide.

We build AI agents that do the slow parts of the job — reading enquiries, drafting replies and quotes, preparing follow-ups — and then stop. A person on your team approves every action that matters, and every step is logged.

How it works

The agent prepares. Your team approves.

Every agent we build follows the same four steps. Nothing that reaches a customer, moves money or changes records happens without a person saying yes.

  1. 01 · AGENT

    Reads

    A WhatsApp enquiry, an email, a form or a document arrives. The agent reads it and pulls the facts it needs.

  2. 02 · AGENT

    Drafts

    It drafts the reply, quote or action — and writes down why, with the sources it used.

  3. 03 · PERSON

    Approves

    Your team sees an approval card: approve, edit or decline. High-risk actions need two different people.

  4. 04 · SYSTEM

    Acts & logs

    Only then is the message sent or the record changed. Who approved, what and when is saved in the activity trail.

Inside the system

What your team works with.

The screens below are illustrations of our real approval flows, redrawn in our brand with sample data. Ask us for a live demo to see the working system.

01 · Approvals

An approval card for every important action

The agent never sends a quote, gives a discount or changes a booking by itself. It puts a card in the approval queue with the draft, the reason and the numbers. One tap to approve, or edit it first.

Risky actions — discounts, refunds, deletions — need two different approvers, and nobody can approve their own request.

02 · Workflows

Step-by-step runs you can follow

Each job runs as a workflow you can watch: read the enquiry, match it to a service, draft the quote, wait for approval, send, schedule the follow-up. If a step needs a person, the run pauses right there.

03 · Activity

A full trail of who did what

Every run keeps its trace: what came in, what the agent drafted, who approved it and what was sent. When a customer asks "who promised this?", the answer is one click away.

04 · Customer chat

Answers customers, books the team

A chat agent answers questions from your own information — services, areas, process — and never invents prices. When a customer is ready, it takes a meeting or site-visit request for your team to confirm.

The manager view shows what the agent used for each answer, so you can check it.

Safety by design

Rules the agent can't skip.

Everyday

One approval

Replies, reminders and quotes go out after one person on your team approves the card.

Risky

Two different approvers

Discounts, refunds and deletions need two separate people to say yes.

Always

No self-approval, full trail

Nobody approves their own request, and every decision is saved with who, what and when.

Examples we can build

Where an approved agent saves hours.

Ideas for Indian businesses like the ones we already work with. Each one drafts the work and waits for your team.

Schools & colleges

  • Fee reminders drafted for each overdue family
  • Parent questions answered from the school calendar and circulars
  • Admission enquiries followed up the same day

Restaurants

  • Review replies drafted for the owner to approve
  • Daily sales summary on WhatsApp at closing
  • Stock reorder lists prepared from the day's sales

Retail & showrooms

  • Khata reminders for customers with dues
  • Festival offers drafted per customer group
  • Purchase orders prepared for low-stock items

Signage & field services

  • Enquiry → quote drafts from photos and sizes
  • Site visits scheduled and confirmed
  • Install reports with photos for each outlet
We use it ourselves

Our own front-desk agent is next on this site.

It answers questions about our services, never quotes prices in chat, and takes a meeting request for our team to confirm. Want the same for your business? Let's show you.

Book a live demo