A WhatsApp AI agent for your business — set up without a developer

An AI agent on WhatsApp answers customer questions in your tone, at 2 am, without a menu of buttons. Setting one up no longer requires a developer — it requires a knowledge base, a short training period, and realistic expectations.

Button bot vs AI agent: not the same thing

A classic chatbot is a script. You draw a flow — "press 1 for prices, press 2 for delivery" — and the bot walks people through it. It is predictable, cheap, and completely lost the moment a customer types "do you have the blue one in medium, and can it arrive by Friday?"

An AI agent reads that free-form question, pulls the relevant facts from your knowledge base, and writes an answer in plain language. No menu, no "sorry, I didn't understand that." Plenty of businesses run both: a keyword bot for the three questions everyone asks, and an AI agent for everything else.

	Button / keyword bot	AI agent
Input it handles	Exact keywords, menu choices	Free-form questions, typos, slang
Setup effort	Draw the flow once	Build a knowledge base, then train the tone
Odd questions	Dead end or fallback message	Usually a reasonable answer
Predictability	Total — it only says what you wrote	High but not total; needs review early on
Running cost	Included in the plan	Pay per token used
Best for	FAQs, routing, opt-in confirmations	Sales questions, objections, after-hours support

Where an AI agent earns its keep — and where it fails

AI agents are genuinely good at a specific set of jobs:

• After-hours questions. An agent that answers at 2 am keeps the lead warm until morning.
• Oddly phrased questions. Real customers type paragraphs with three questions inside, not keywords. AI parses that fine.
• First-line objections. "Why is it pricier than the other shop?" has a good answer in your docs — the agent gives it instantly instead of losing the lead to silence.
• The same twenty questions, forever. Delivery times, sizes, return policy. Boring for you, effortless for the agent.

And there are places where handing the conversation to AI is a mistake. Payments and refunds — anything where a wrong answer costs real money — should go to a human. So should regulated topics (medical, legal, financial), because an agent answers confidently even when it should not. And angry customers: an upset person wants a person, and a smooth AI reply tends to make it worse.

Setup in chat-valet: knowledge base, then tone

1. 1
   Connect your account

   The AI assistant works on personal WhatsApp, Telegram and MAX accounts — no Business API contract, no developer setup.

2. 2
   Upload a knowledge base

   Product docs, price list, delivery terms, the FAQ you keep copy-pasting from. The agent answers only from what you give it — a thin knowledge base means thin answers.

3. 3
   Describe your tone of voice

   Short and friendly or formal, emoji or not. The goal is replies that read like you, not a press release. More in how to make an AI assistant sound human.

4. 4
   Pick a work mode

   Auto-reply, draft for approval, or ask first. Start conservative — more on this below.

Draft mode is your training wheels

The three modes differ in how much the agent is allowed to do on its own. In auto mode it replies to customers directly. In draft mode it writes the answer but waits for your approval — you get the draft in a Telegram bot, read it, and send or edit it. In ask first mode it checks with you before engaging at all.

Our honest advice: do not start in auto mode. Run the agent in draft mode for a week or two. You will see exactly what it would have said, catch answers that are technically correct but phrased wrong, and fix the knowledge-base gaps that caused them. Approving a draft takes ten seconds from your phone; cleaning up after a bad automated reply takes much longer.

When the drafts stop needing edits — usually once the knowledge base covers the real questions, not the ones you guessed — switch routine topics to auto and keep sensitive ones on approval.

Escalation: the agent should know when to stop

A good AI agent is not one that answers everything — it answers the routine 70% and hands the rest to you cleanly. In chat-valet every conversation, AI-handled or not, sits in one unified inbox covering WhatsApp, Telegram and MAX (a Russian messenger with roughly 85 million users, according to public estimates; outside Russia, WhatsApp and Telegram alone work fine). You can step into any chat at any moment, and the customer never repeats themselves.

Practical escalation setup: keep payments, refunds and complaints on human handling; let auto-replies and keyword bots acknowledge messages instantly outside working hours; and use quiet hours so nobody gets a message at 3 am local time.

WhatsApp AI agent price: what it actually costs

"WhatsApp AI agent price" is one of the most-searched questions on this topic, and most vendors answer it with "book a demo." Here is the arithmetic instead. In chat-valet the cost has two parts: the plan — from $7/month — and the AI's actual token usage. A typical short customer reply costs a fraction of a cent; a hundred AI-answered conversations a month usually add up to less than a cup of coffee.

You choose the model tier yourself: a cheaper model for simple FAQ traffic, a stronger one for genuinely complex questions. No per-seat AI fee, no fixed per-conversation charge — a slow month costs almost nothing beyond the plan. The 30-day free trial needs no card; AI usage is the one thing not included, since it is metered from the first token.

The honest part: personal accounts and risk

chat-valet works through personal messenger accounts, not the official WhatsApp Business API. That keeps it cheap and simple, but there is no platform-sanctioned safety net, and nobody — us included — can guarantee an account will never be flagged.

The good news for AI-agent use specifically: answering people who messaged you first is the lowest-risk thing a personal account can do. The risk lives on the outbound side. If you also send broadcasts from the same number, keep them to 20–30 messages a day, use the built-in random 30–90 second delays, warm up new numbers slowly, and send only to people who opted in — in the EU and many other markets that is a GDPR and consent requirement, not just a safety tip. See how to build a contact list legally before any cold outreach.

Connect your WhatsApp, upload your FAQ, and watch the first drafts come in — free for 30 days, no card required.

Start free trial

Frequently asked questions

Do I need a developer to set up a WhatsApp AI agent?

Not with chat-valet. You connect your WhatsApp account, upload your docs as a knowledge base, describe your tone, and pick a work mode. A REST API and webhooks exist for deeper integration, but the basic setup is point-and-click.

How much does a WhatsApp AI agent cost?

Two things: a plan from $7/month plus the AI's actual token usage. A typical short reply costs a fraction of a cent, and you choose the model tier yourself. No per-seat fees, no fixed per-conversation charge.

What happens when the AI doesn't know the answer?

You decide. In "ask first" mode it checks with you before replying; in draft mode every answer waits for approval. Conversations can always go to a human — everything lands in one unified inbox, so picking up a chat mid-conversation is normal.

Can the AI really sound like me and not like a generic chatbot?

It answers from your knowledge base in a tone you describe — short and informal, formal and detailed, whatever fits. It will not be perfect on day one; that is what draft mode is for. Correct its phrasing for a week or two and it gets close to how you actually write.

Is it safe to run an AI agent on a personal WhatsApp account?

Replying to people who message you first is far lower-risk than mass outbound sending, but it is still a personal account, and no tool can guarantee it will never be flagged. Keep broadcasts to 20–30 messages a day per number, warm up new numbers, and message only people who opted in.

Is the AI assistant included in the free trial?

The 30-day trial (no card required) covers broadcasts, bots and the unified inbox. AI usage is not included — it is billed per token from the start, because the cost depends on how much you use it.