Platform · Agent studio

Configure an agent's knowledge, actions and limits, then test it like a new hire.

Name: AksharAI
Brand: AksharAI
Price: 7500 INR

Agent studio is where you set up an agent: what it knows, what it may do, how it should sound and when it must stop and ask. You test it on real questions in a sandbox and release it only when your team is happy with the answers.

Book a demo See the agents

AksharAI / record

AksharAIAgent: WhatsApp supportSearchAK

Agent: WhatsApp support

Channel: WhatsApp Business, English and Hindi
Knowledge: Returns policy v4, product catalogue, FAQ sheet
Allowed actions: Look up order, raise return request, create ticket
Blocked: Refund approval, discounts, skin or medical advice
Hand over to: Support desk, 9 am to 9 pm
Version: v7, published 3 Oct 2026

Knowledge sources linked
Output checks added
Tested on 300 past chats
Live on 25% of new chats
Full release

Representative screen with sample data. Client screens stay private.

The problem

A support bot that answers confidently and wrongly

A D2C skincare brand gets hundreds of WhatsApp messages a day: where is my order, can I return an opened product, which serum suits oily skin. A generic chatbot tried last year made up a return window and promised refunds the brand does not offer. The support team switched it off and went back to answering every message by hand, including the same order status question forty times a day.

What it gives you

Inside agent studio

Prompt flow design

Set the agent's role, tone and the order it works in: greet, identify the customer, understand the question, answer or hand over.

Knowledge sources

Point the agent at the documents and data it may use, such as the returns policy, product catalogue and order system, and nothing else. Answers cite the source they came from.

Triggers and actions

Decide what starts the agent, such as a new WhatsApp message or an email to support, and which actions it may take, such as looking up an order or raising a ticket.

Output checks

Check each answer before it is sent: no promises outside policy, no medical claims, required details present. Answers that fail go to a person.

Visual editor and testing

See the agent's setup on one screen and run it against a bank of real past questions. Compare answers across versions before you release a change.

Workspace governance and deployment controls

Control who can edit, test or publish an agent, keep a version history, and roll back to the last good version in one step.

How it works

The order you work in

Step 1Give it a jobWrite down what the agent handles, what it never handles and who it hands over to.
Step 2Connect what it may knowAdd policies, FAQs and read access to the systems it needs, such as the order database.
Step 3Set its actions and limitsAllow specific actions, block everything else, and add the output checks your team cares about.
Step 4Test on real questionsRun last month's customer messages through it. Your support lead marks each answer right, wrong or needs a person.
Step 5Release in stagesStart with one channel or a share of traffic, read the transcripts, then widen it.

AksharAI / assistant

AksharAITest consoleSearchAK

Test console

Hi, my order AR-58213 still says shipped. When will it reach Pune?
Your order left the Bhiwandi warehouse on 6 Oct and the courier shows delivery by 9 Oct. I will share the tracking link here.
Also the face wash gave me a rash. Can I get a refund?
Output check: skin reaction mentioned. Handing over to support desk.
I am sorry to hear that. I have passed this to our support team with your order details, and a person will reply here shortly.

Representative screen with sample data. Client screens stay private.

An example

WhatsApp support for a D2C skincare brand

The agent answers order status, delivery dates and returns questions using the brand's policy document and read only access to the order system. It may raise a return request but never approve a refund. Questions about skin conditions or reactions always go to a trained support person, with the chat so far attached. The support lead reviews a sample of transcripts every week and adds new questions to the test bank when the agent gets one wrong.

An illustrative example with sample data.

Questions

About agent studio

How do we stop an agent from making things up?

Limit what it may use to the sources you connect, ask it to cite them, and add output checks that block answers outside policy. Questions it cannot answer from those sources go to a person. Test banks built from real past questions show you where it still slips.

Who decides when an agent is ready?

Your team. Agent studio shows test results version by version, but publishing is a permission you give to named people, usually the head of the team the agent serves.

Can one agent work on several channels?

Yes. The same agent can answer on WhatsApp, email, a web widget or inside an internal app, with channel specific settings such as message length or working hours.

Keep going

Bring one slow, repetitive process. We will show you an agent doing it.

A working session with someone who builds these every week.