Replace your help desk with an AGNT support agent

The knowledge base is the foundation of every support interaction. The agent can only answer what it knows, and it knows what you put in the knowledge base. Upload your existing help documentation — FAQ pages, product manuals, policy documents, troubleshooting guides — as PDF, DOCX, or plain text files.

AGNT chunks each document into overlapping segments, embeds them, and tags them by type (FAQ, troubleshooting, policy, billing, product). When a customer asks a question, the agent retrieves the most relevant chunks and constructs an answer grounded in your documentation — not in general-purpose training data.

Start with your top 20 support tickets from the last month. If the knowledge base can answer those 20 questions correctly, you are ready to go live. If any answers are wrong or incomplete, add the missing documentation before proceeding.

Connect your support channel in the AGNT dashboard. For WhatsApp, use the 360Dialog integration (same setup as venue agents — authorize your WhatsApp Business number via OAuth). For Telegram, create a bot via BotFather and paste the bot token into AGNT.

You can connect multiple channels simultaneously. The agent maintains conversation context per customer across channels — if a customer starts on WhatsApp and switches to Telegram, the agent has the full history.

Configure your welcome message and business hours. Outside business hours, the agent still responds (it does not sleep), but it can adjust its tone: 'I can help with most questions right now. If this needs a human, I will escalate and someone will follow up by 9 AM.'

Define ticket categories and priority levels. The agent classifies every inbound message by category (billing, technical, product question, complaint, feature request) and priority (P0 urgent, P1 high, P2 normal, P3 low) based on content analysis and customer context.

Configure routing rules per category: billing questions go to the finance team, technical issues to engineering support, complaints to the customer success manager. The agent applies these rules in real time — no manual triage queue, no ticket sitting unclassified for hours.

For P0 issues (service outage reports, security concerns, payment failures), the agent escalates immediately to the on-call human via the configured alert channel (Slack, PagerDuty, WhatsApp) regardless of the normal flow.

The agent resolves FAQ-type questions directly from the knowledge base. 'How do I reset my password?', 'What is your refund policy?', 'Do you support multi-currency?' — these are answered in seconds with grounded, accurate responses pulled from your documentation.

Unlike a static FAQ page, the agent handles follow-up questions in context. If a customer asks about the refund policy and then asks 'Does that apply to annual plans too?', the agent understands 'that' refers to the refund policy and responds accordingly.

Track FAQ resolution rates in the AGNT dashboard. The agent reports which questions it resolves autonomously vs which require human intervention. Use this data to identify knowledge base gaps — every escalated FAQ is a missing document.

Not every support interaction should be automated. Configure escalation triggers: customer explicitly asks for a human, sentiment drops below a threshold (frustration detection), the agent's confidence score is below 70%, the conversation exceeds a turn limit without resolution, or the topic matches an escalation keyword list.

When an escalation triggers, the agent hands off seamlessly. The human receives: customer name and account details, the full conversation transcript, the agent's classification (category, priority), the agent's attempted resolution (if any), and a suggested next action.

The customer sees: 'I am connecting you with a team member who can help with this. They will have our full conversation, so you will not need to repeat anything.' No dead air, no queue position number, no hold music.

The AGNT dashboard provides support metrics in real time: first response time, resolution time, auto-resolution rate, escalation rate, customer satisfaction (collected via post-conversation survey), and volume by category.

The metrics that matter most: auto-resolution rate (target: 70%+ within the first month) and escalation-to-resolution time (how quickly humans close what the agent could not). If your auto-resolution rate is below 60%, the knowledge base needs more content. If escalation resolution time is high, your routing rules need adjustment.

The agent also generates a weekly support digest: top questions asked, new questions the knowledge base does not cover, resolution rate trends, and recommended knowledge base additions. This creates a flywheel — the more the agent runs, the better the knowledge base becomes, and the higher the auto-resolution rate climbs.