How to set up and run your Safqat AI agent

This guide explains the logic behind a solid setup, the order of settings, and how to move from teaching and preparation to internal testing and then launch on WhatsApp or the web widget.

Start by teaching the AIFeed it accurate dataTest internally before go-liveImprove continuously

Core idea

Treat the agent like a digital teammate — not a single “on” switch. It needs clear, structured content first so answers stay high quality. You begin with what you already have: products, instructions, stable facts, FAQs, links, images, and videos where relevant. Then you run it quietly (not announced to the public), review it, and test for several days in the simulator. When results look good, you can go live on WhatsApp or embed the web widget on your site.

Teams that later need richer analytics, voice, and higher message volume usually outgrow the basics — Safqat Pro is built for that stage.

Onboarding path: from zero to launch

1) Gather data

Products / facts / FAQs

2) Teach the agent

Instructions + knowledge + policies

3) Quiet run

No public launch yet

4) Test & refine

In the simulator for several days

5) Go live

WhatsApp / web widget

Rule: don’t launch the agent until you like the results in testing.
Step 1

Subscribe and activate your plan

Use the plan page to subscribe securely. After checkout, your workspace opens so you can configure the agent. Bookmark your dashboard — that’s where you’ll spend most of your setup time.

Practical notes

  • Use an active email so billing and important alerts reach you.
  • After successful payment, the main dashboard is where you manage the agent.

Payment is processed securely; your plan appears under billing once active.

Step 2

Understand the main dashboard

The dashboard is your control center: agent status, conversation volume, and engagement. Turning the agent “on” starts auto-replies on the connected channel — keep it off until setup feels ready.

Operational rule

  • Agent off = you can still reply manually on WhatsApp.
  • Agent on = automatic replies start on the linked channel.
Step 3

Agent instructions: behavior and rules

Here you define the agent’s personality and boundaries: tone, what it should never promise, and how it handles edge cases. Keep volatile numbers (prices that change often) in products or the knowledge base instead of hard-coding them only in this box.

Practical notes

  • Write style, conditions, and guardrails for every conversation.
  • You can add a few scenario examples — but don’t limit the whole section to one narrow case.
  • Don’t rely on “we’ll fix it live” — validate in the simulator first.
  • When you’re ready for keyword-based multi-step WhatsApp flows (text, image, voice), that’s a Pro capability worth exploring.
Step 4

“I can’t answer that” message

Configure a clear message for when the agent shouldn’t invent an answer. This isn’t a failure — it’s responsible automation. When information is missing, saying so beats guessing.

Example

Sorry — I’m not sure I understood. Could you share a bit more detail?

Principle

  • Unknown is better than wrong.
  • These moments become your backlog for new knowledge and better coverage.
Step 5

Knowledge base: facts and reference material

The knowledge base is where you store what the agent should rely on when answering: stable facts, FAQs, important links, service descriptions, and supporting images, video, or documents.

Best practices

  • One clear entry per fact or policy.
  • Use meaningful titles: payments, cancellation, hours, booking links.
  • You can type or dictate updates to move faster.
Step 6

Products: what you sell

Add the concrete products or services you offer — e-commerce SKUs, listings, packages, or anything customers should browse or ask about. The agent uses this to present options accurately.

Step 7

Handoff to your human team

Early on, the agent won’t know everything. Set a calm handoff message for cases that need a person — unclear requests, sensitive topics, or missing knowledge.

Example

Thanks — our team will follow up with you here shortly. Leave your phone number if you’d like a callback.

Practical notes

  • At the start, prefer escalation over risky guesses.
  • Repeated escalations usually mean a knowledge gap to fix.
  • As knowledge improves, escalations naturally go down.
Step 8

Feedback loop and continuous improvement

After the agent is live on WhatsApp or the web, watch what it misses. Turn unanswered or weak answers into knowledge updates. This is how quality compounds week over week.

Step 9

Final testing in the simulator

Before exposing the agent to everyone, test inside the dashboard simulator. Play the customer: basic questions, hard questions, booking flows, and escalation paths.

What to validate

  • Tone and style match your brand.
  • The agent asks for the right details at the right time.
  • Unknown cases don’t turn into confident mistakes.

When you’re happy with the results

After several days of review — not a single rushed afternoon — you’re in a good place to enable the agent for real traffic on WhatsApp or the web widget.

Link for this step

Optional

Welcome-only mode

Some teams start with a welcome message and immediate routing to a human for all Q&A. It’s a valid transitional step while you finish knowledge and instructions.

When it helps

  • You want instant acknowledgment without giving the AI full responsibility yet.
  • Your team still owns complex questions until you’re confident in coverage.

When you’re ready for more than the basics: Safqat Pro

Room to scale

A much higher monthly AI message allowance so busy inboxes don’t hit the ceiling during campaigns or peak season.

Richer customer journeys

Advanced WhatsApp flows with media — structured paths that feel closer to how your team already sells.

Voice that matches your market

Voice understanding and voice replies that fit real customer behavior — not just text.

Concepts (signals you can act on)

Pro helps you spot meaningful moments inside a chat — so your team gets notified when something important happens, not after the lead goes cold.

Concrete example

A customer shares address and phone for delivery — the system can recognize that pattern and alert your team on WhatsApp or email for immediate follow-up.

If your funnel is growing and you care about depth, not just speed, compare Pro side-by-side with your current limits.

View Safqat Pro

Practical summary

Start here
  • Add products/services and their links.
  • Enter facts, policies, and knowledge.
  • Tune instructions and escalation.
  • Test daily in the simulator.
Then go live
  • Enable the agent when answers feel ready.
  • Embed the web widget where it matters.
  • Keep reviewing real conversations and improving.
Agent onboarding guide — configure Safqat step by step | Safqat AI