> ai / ai-agents
AI Agents
An agent is more than a chatbot. It reads a request, looks up your data, and does the work, such as creating an order, checking stock or sending a quotation, with a person approving the steps that matter.
> capabilities
What you get
Agents on WhatsApp and the web that take real actions in your ERP or POS.
WhatsApp and web channels
Customers and staff talk to the agent where they already are, in WhatsApp or a chat widget on your site.
Actions in your systems
The agent calls your ERP, POS or CRM through APIs to read records and create or update them.
Approvals and limits
Set what the agent can do alone and what needs a human to confirm, such as refunds or large orders.
Grounded answers
Replies draw on your price lists, policies and documents, and the agent says when it does not know.
Handover to people
When a case is unclear, the conversation passes to your team with the full history attached.
Logs and review
Every conversation and action is recorded so you can audit, correct and improve behaviour.
> use cases
Where it fits
- Order taking
- Customers send an order by message and the agent creates it in your ERP after checking stock and prices.
- Internal assistant
- Staff ask for sales totals, stock levels or pending invoices in plain language.
- Support triage
- The agent answers common questions and routes the rest to the right person.
- Follow-ups
- Payment reminders and delivery updates sent and tracked without manual effort.
> process
How it runs
- 01
Pick one workflow
Choose a task that is frequent, rule-based and easy to measure.
- 02
Connect and define limits
Wire up the systems the agent needs and agree what it may do alone.
- 03
Test on real conversations
Run it against past messages and a small live group before wider rollout.
- 04
Launch and tune
Review logs weekly, fix mistakes and add the next task.
> faq
Common questions
Can the agent work with our existing ERP or POS?
Yes, if it has an API or database access we can use. It also works natively with SahaAI, SelloKart and Order Pilot.
What if the agent gets something wrong?
Sensitive actions require approval, everything is logged, and we add tests for each mistake we find.
Which languages are supported?
Modern language models handle many languages. We test the ones your customers actually use.
> let's build something
Tell us where you want to go with AI, data or cloud — we'll reply within one working day.
