How to Configure 24/7 AI Customer Service in WhatsApp: Personality, Knowledge, Skills, and Automations

WhatsApp can be the 1-to-1 entry point where customers talk to Hermes Agent, ask common questions, and receive service drafts or routing guidance.

Here, 24/7 means that the Agent can respond whenever the service is running, the model is available, and the channel connection is healthy; it is not an uninterrupted-service promise or a human-support SLA.

WhatsApp proactive messages are still affected by conversation windows, message templates, and channel configuration, so automated reminders must be designed separately from customer-initiated conversations.

Summary

WhatsApp can serve as a one-to-one entry point between customers and a Hermes Agent, useful for answering recurring questions, capturing customer needs, and drafting responses. But "24/7" here means it can respond whenever the service is running, the model is available, and the channel is connected — it is not an uninterrupted-service guarantee or a human-support SLA. A second constraint matters just as much: proactive messaging on WhatsApp remains subject to conversation windows, template messages, and channel configuration, so automated reminders must be designed separately from customer-initiated conversations rather than assuming you can push at will. Build in four layers, in order: the persona sets voice and escalation conditions, knowledge captures FAQs, skills add reusable procedures, and automations handle scheduled summaries.

Who this is for

This guide is for people who want to deploy a personal support agent, product-information assistant, or after-sales triage assistant in WhatsApp.

It is especially useful for teams that need consistent support language, maintainable FAQs, fewer repetitive replies, and a clear handoff path for complex requests.

Prerequisites

A running LightVela Hermes Agent. If you have not created one yet, start with the Quickstart.
A working model configuration or available trial credits.
A usable WhatsApp account connected by following Configure Channel - WhatsApp.
Product material, service hours, escalation paths, and human contact details.
A clear definition of actions that require human execution or confirmation.

Understand the five layers of 24/7 customer service

Layer	Responsibility	Suitable location
Personality	Role, voice, priorities, and service boundaries	SOUL.md or Personality
Knowledge	Product FAQs, policies, service hours, and escalation paths	Reference files in Cloud Storage
Skills	Retrieval, organization, drafting, and other concrete actions	Skills
Memory	Confirmed customer preferences and durable collaboration context	Memory Management
Automations	Scheduled checks, internal reminders, and proactive notices	Automations

Do not put a large product knowledge base into the personality, and do not treat unconfirmed customer content as long-term memory.

Steps

1. Connect and verify a WhatsApp 1-to-1 conversation

In the target Hermes Agent's channel settings, select WhatsApp and complete QR authorization by following Configure Channel - WhatsApp.

Open the corresponding contact in WhatsApp and send a test message, then confirm that the console shows the channel as connected and that the reply arrives normally.

The current WhatsApp channel does not support group chats, so this customer-service flow is designed for 1-to-1 customer conversations.

2. Write a customer-service personality brief

Use the SOUL.md personality guide to define the support assistant's identity, voice, response order, and escalation boundary.

Start with a brief like this:

You are our product support assistant. Give the customer the most important conclusion first, followed by the necessary steps. Answer only from confirmed product material, and clearly suggest a human handoff when uncertain. Do not promise refunds, price changes, identity checks, data deletion, or other high-risk outcomes. Before sending content to an external system or making a change, explain the impact and wait for human confirmation.

Keep stable service principles in the personality, and keep concrete product answers in maintainable reference files.

3. Organize the support knowledge base

Prepare at least these categories:

Product and plan FAQs.
Service hours, response expectations, and human contact details.
Refund, cancellation, upgrade, and incident policies.
Common errors and troubleshooting steps.
Requests that require human handoff and the information to collect before handoff.

Split the material into maintainable topic files and place them in Cloud Storage.

Test the material with a low-risk prompt:

Answer this question only from the prepared support material. If the material does not contain an answer, say what information is missing and suggest a human handoff.

4. Install and validate the necessary Skills

If you need knowledge retrieval, ticket organization, web lookup, or content generation, choose a trusted Skill from Skills.sh or ClawHub.

Before asking Hermes to install it through WhatsApp, ask what the Skill reads, which external services it calls, and whether it can send proactive messages.

High-risk Skills may request secondary confirmation before installation, so review the permissions and safety guidance in Configure Skills.

After installation, test drafts, summaries, and FAQ retrieval before allowing any Skill to process refunds, delete data, or send unreviewed customer content.

5. Set customer preferences and durable memory

Only information that is confirmed and useful for future service should be saved to memory or the user profile.

Examples include a confirmed preference for English, a preference for conclusion-first replies, or the status of an ongoing support case that the customer has verified.

Passwords, verification codes, payment details, full identity information, and unnecessary sensitive data should not be saved as long-term memory.

Ask Hermes in conversation to remember or forget a preference, and periodically review the Memory Management page in the console.

6. Configure automations and proactive messages

Good automation candidates include internal task reminders, unresolved-case checks, and daily service summaries.

Configure the schedule, active time window, and notification method in Automations.

WhatsApp proactive messages may be limited by conversation windows or approved message templates, so do not assume that a normal text can always be delivered when the customer has not started a conversation.

Use a test number first to verify delivery, content, and the internal handling path when a notification fails.

7. Establish human handoff

When a request involves refunds, payments, identity checks, legal commitments, escalated complaints, or missing reference material, the support Agent should stop guessing and hand off to a human.

The handoff message should include the known facts, missing information, recommended next step, and whether the customer needs to repeat the issue.

Review a sample of conversations each day, focusing on wrong citations, overpromising, sensitive-data handling, and repeated proactive messages.

A reusable support response order

Give the direct conclusion.
Cite the most relevant material or steps.
Explain constraints, cost, timing, or required information.
State clearly when a human handoff is needed.
Execute an external action only when it is allowed and confirmed.

Configure Channel - WhatsApp - Complete the WhatsApp 1-to-1 connection.
How to Turn an IM Agent into a Customer-Service, Community, or Companion Assistant with Skills - Choose and validate customer-service Skills.
How to Define an IM Agent's Role, Personality, and Voice with SOUL.md - Define the support personality and behavioral boundary.
Automations - Configure internal reminders and scheduled summaries.