Your customers don't stop having questions at 5 PM. They message on weekends, late at night, and across time zones you're not covering. A human support team can't be online around the clock without serious cost. An ai whatsapp bot that actually knows your business — your pricing, your return policy, your troubleshooting steps — can be.
The Blueticks AI Support Agent is built on top of the Blueticks Bot System. It combines a knowledge base of your documents with an Agent Node that reads incoming WhatsApp messages, searches that knowledge base, and sends back grounded answers. No hallucinated policies. No generic deflections. The answers come from what you uploaded.
This article walks through how the system works, how to set it up, when it escalates to a human, and what it genuinely cannot do. If you're evaluating this for a business, you need the honest version — that's what follows.
What the Agent Actually Does — The Full User Flow
When a customer sends a message to your WhatsApp number, the Trigger Node activates the bot flow. The message passes to the Agent Node, which searches your knowledge base for relevant content, formulates a response using that content, and sends the reply — typically within seconds.
From the customer's side, it reads like a normal WhatsApp conversation. There's no "chatbot flavor" indicator in the UI. The response style is whatever you configured: formal, casual, somewhere in between.
From your side, you get:
- Automated responses to common product questions, troubleshooting queries, pricing requests, and policy lookups
- A dashboard showing the number of conversations handled, average response time, common questions, and how often the bot hands off to a human
- Conversation transcripts you can review to improve the knowledge base over time
The bot operates through the Chrome extension running alongside WhatsApp Web. That means it requires your browser session to be active — or you to be on the Pro plan with the offline gateway enabled, which handles delivery when your machine is off.
One important design note: the Agent Node is an endpoint in the flow, not a pass-through. Once the agent is handling a conversation, it operates autonomously. It doesn't exit to other nodes. If it can't answer something, it follows your configured handoff rules — more on that below.
For a full walkthrough of the feature, see the AI Support Agent guide.
How It Answers — Knowledge Base Ingestion and Multilingual Support
The agent's answer quality is directly proportional to what's in the knowledge base. This is worth repeating because it's the biggest variable in whether the feature works for you.
The knowledge base accepts uploaded documents. Supported formats, per the knowledge base documentation:
- PDF (
.pdf) - Microsoft Word (
.doc,.docx) - HTML files (
.html,.htm) - Markdown (
.md,.markdown) - Plain text (
.txt)
When you upload a file, the system extracts the text, chunks it, and indexes it using semantic embeddings. Documents are searchable within minutes. When a customer asks a question, the agent queries the index, ranks chunks by relevance, and uses the top results to formulate the response.
What this means practically: If your product catalog is in a PDF and a customer asks "do you ship to Germany?", the agent finds the relevant section and answers based on it. If shipping policy isn't in any document, the agent will say it doesn't know — which is the correct behavior. You can configure a fallback message for unanswered questions.
A note on URL ingestion: The setup wizard in the product mentions website URLs as a potential knowledge base source. The underlying knowledge base documentation is explicit that URL crawling is not currently supported — only file uploads work. If your content lives on a website, export it to PDF or HTML and upload that file.
Multilingual support works at the document level. The agent can respond in the language the customer writes in, but answer quality improves when the underlying knowledge base includes documents in that language. If your KB is English-only and a customer writes in Spanish, the agent will attempt to answer from English sources — with varying accuracy. For businesses serving non-English markets, uploading translated documentation produces noticeably better results.
Setting It Up — Step by Step
Setup takes three meaningful phases: installation, knowledge base creation, and agent configuration. Here's the sequence.
Phase 1: Install and Activate
- Install the Blueticks Chrome extension if you haven't already.
- Open WhatsApp Web in Chrome.
- Click the Blueticks icon in your browser toolbar.
- Select AI Support Agent from the menu.
- Toggle the agent to Active.
Phase 2: Build the Knowledge Base
From the AI Support Agent dashboard, go to Knowledge Base and upload your documents. Practical starting point: a FAQ document, your product or service catalog, and any pricing or policy page you'd normally direct customers to.
The docs recommend starting with the most frequently needed content rather than trying to upload everything at once. Upload, run a test query using the Test Query feature on each document, and verify the answers look right before going live.
A few guidelines from the documentation that hold up in practice:
- Use clear, direct language in your documents. Well-structured docs produce better retrieval.
- Organize content into separate knowledge bases by topic if you have distinct product lines or service areas.
- Large documents are automatically chunked, so you don't need to pre-split them — but very dense technical documents may require more precise customer questions to surface the right section.
Keep your knowledge base current. If your pricing changes and the old document is still indexed, the agent will quote the old price.
Phase 3: Configure Agent Behavior
In Agent Settings, configure:
- Greeting message — what the agent says when a conversation starts
- Working hours — when to engage vs. when to send a "we'll get back to you" message
- Handoff triggers — the conditions under which the conversation transfers to a human agent
- Response style — formal or casual tone
- Languages — which languages the agent should respond in
The Agent Node also lets you choose from pre-built system prompt templates: Customer Support (General FAQs), Product or Service Information Assistant, Returns/Warranty/Policy Info Assistant, and others. Pick the one closest to your use case and customize it with your company name and any specific constraints.
The system prompt is not visible to customers, but it shapes every response. Worth spending time on. If you tell the agent "do not speculate on delivery dates," it won't.
Phase 4: Test Before Going Live
Use the Test Agent feature to run simulated customer queries before the agent handles real conversations. Then conduct a live test from a second WhatsApp account. Verify that edge cases — unusual questions, questions with no answer in the KB, multi-part questions — produce reasonable responses or appropriate handoffs.
The Blueticks whatsapp support automation documentation recommends starting with an Allow List (restricting access to your own number) in the Trigger Node settings during testing, then opening access once you're satisfied.
When It Escalates — Human Handoff Logic
The Agent Node is designed with a clear principle: complex, nuanced issues require human intervention, and the system should route them there without friction.
You configure handoff triggers in Agent Settings. These are conditions — a customer asking for a refund, expressing frustration, or asking a question the agent flags as outside its scope — that transfer the conversation to a human agent. The transition is meant to be smooth: the human picks up the conversation with context from what was already exchanged.
How you set handoff triggers determines how well this works. Too loose and the agent hands off constantly, defeating the automation. Too strict and customers with legitimate complex needs wait for a bot that can't help them.
The documentation recommends having human agents available for handoffs even while the AI Support Agent is running. The bot handles volume. Humans handle exceptions. That's the division of labor the system is designed for.
One configuration worth noting: the Trigger Node supports both block lists and allow lists, so you can exclude specific contacts from hitting the bot at all — for instance, high-value clients you always want to handle personally.
A best practice called out explicitly in the product docs: disclose to customers that they're talking to an AI. The product doesn't enforce this, but it's the right call for trust and for compliance depending on your market.
What It Doesn't Do — Honest Limits
The AI Support Agent is useful. It's also not a general-purpose AI assistant, and there are gaps worth knowing before you build a workflow around it.
It doesn't know what you haven't uploaded. The agent answers from the knowledge base. If a customer asks something your documents don't cover, the agent either says it doesn't know or triggers a handoff. It doesn't synthesize answers from general web knowledge or make things up — which is actually the behavior you want for a support bot.
It doesn't crawl URLs. As noted above, the knowledge base is file-upload only. No live web crawling, no syncing from a help center URL.
It's browser-dependent on most plans. The Chrome extension requires an active WhatsApp Web session to send responses. On the Free, Basic, and Standard plans, if your machine is off, the bot is off. Automated whatsapp replies through the AI Support Agent require the offline gateway, which is a Pro plan feature.
Multilingual quality varies. The agent can respond in multiple languages, but accuracy depends on having source documents in those languages. English-only knowledge bases serving non-English customers will produce inconsistent results.
Agent Nodes consume credits per tool call. Each time the agent searches the knowledge base or uses another tool, it costs one credit. A single customer interaction may involve multiple tool calls — meaning a multi-step conversation uses more credits than a simple Q&A. Credit allocations by plan: 50 (Basic), 100 (Standard), 200 (Pro). Monitor usage during early deployment to size your plan correctly.
No beta label, but expect iteration. The product docs don't call this a beta feature. That said, any AI system's answer quality is a moving target — it depends on your KB, your system prompt, and ongoing refinement. Budget time for tuning after launch.
Three Real-World Scenarios
These scenarios are based on the documented capabilities and the use cases the product explicitly describes.
Scenario 1: E-commerce product questions
A customer messages at 10 PM asking whether a product is compatible with a specific device. The agent searches the product catalog PDF, finds the compatibility table, and replies with the relevant row. If the catalog doesn't have that specific device listed, the agent says so and offers to connect the customer with a human during business hours.
Scenario 2: Service business FAQ automation
A salon or clinic receives the same questions every day: "What are your hours?", "How do I book?", "What's your cancellation policy?" The agent handles all of these from an FAQ document uploaded to the knowledge base. The human team only sees conversations that escalated — unusual requests, complaints, special bookings.
Scenario 3: Internal team knowledge assistant
Using the same bot system with an Allow List restricted to staff numbers, a team uses the agent to answer internal procedure questions. An employee asks "what's the process for submitting an expense above $500?" The agent finds the relevant section in the uploaded employee handbook and responds. The knowledge base documentation notes internal procedures as one of the explicitly supported use cases.
Pricing and Plans
The AI Support Agent feature is available on paid plans. The relevant cost factor is AI credits — each agent interaction consumes credits based on tool calls made.
| Plan | Price (annual) | AI Credits | Offline Sending | |------|----------------|------------|-----------------| | Free | $0 | 0 | No | | Basic | $7/user/month | 50 | No | | Standard | $20/user/month | 100 | No | | Pro | $45/user/month | 200 | Yes |
For businesses running the AI Support Agent at volume, the credit allocation matters. A conversation that involves three knowledge base lookups costs three credits. On the Standard plan, 100 credits covers roughly 33 such conversations per billing cycle. Heavy usage needs the Pro plan — or monitoring to ensure you're not depleting credits mid-month.
The offline sending requirement is the other meaningful constraint. If you want the bot to respond to messages that arrive while your browser is closed, that's the Pro plan. For businesses where someone is always at a desktop anyway, Standard or Basic may be sufficient.
All plans include a 30-day money-back guarantee. Pricing is per user, so multi-seat teams multiply accordingly.
Install the Blueticks extension to get started.
Frequently Asked Questions
Does the Blueticks AI Support Agent work with WhatsApp Business accounts?
Yes. The bot system operates through WhatsApp Web, which supports both personal WhatsApp and WhatsApp Business accounts. Setup is identical for both.
Can I run the AI Support Agent on multiple WhatsApp numbers?
The system is per-user, and each user connects one WhatsApp account. For multiple numbers, you'd need multiple seats. Pricing scales per user on all plans.
What happens to customer messages — are they stored on Blueticks servers?
Per the product's privacy documentation, messages, numbers, contacts, and chats are not stored on Blueticks servers. Knowledge base documents are encrypted. Review the full privacy policy for specifics applicable to your jurisdiction.
How many documents can I upload to the knowledge base?
The documentation notes that knowledge base storage limits apply based on your plan, but specific per-plan limits aren't published in the current docs. Contact support at support@blueticks.co for current limits before committing to a large KB migration.
What if the agent can't answer a question?
The agent is configured to say it doesn't know and trigger a handoff rather than generate a speculative answer. This behavior is reinforced by the system prompt — explicitly telling the agent not to speculate produces consistent results. The example system prompt in the documentation reads: "if you don't know, say you'll forward the request to a human."
Is the AI Support Agent available on mobile?
The Blueticks extension runs in Chrome on desktop. WhatsApp Web mobile browser support is limited. For automated replies from a mobile device, the workflow requires routing through a desktop Chrome session.