The Surface Parity Friction: Why disparate logic for web and phone support creates knowledge silos

A customer starts a conversation on your website to check an order status. They receive a partial answer but need to leave for a meeting. Ten minutes later, they follow up on WhatsApp. If your WhatsApp agent uses a different knowledge base or has different API permissions than your web agent, the customer has to repeat their order number. This context reset is where leads abandon the funnel. This failure creates The Conversation Persistence Tax, where losing customer history resets the entire resolution clock.

Many companies try to solve multi-channel support by building separate bots for different platforms. They use a specific widget builder for the website and a different automation provider for their WhatsApp Business account. This approach creates a technical maintenance burden. You end up managing separate sets of prompts and different knowledge libraries for every channel. When you update a shipping policy on your website, you have to manually sync that change to every other bot in your stack. This often leads to The Platform Pivot Penalty because forcing a customer to switch channels often results in a lost lead.

Duvi removes this divergence by using one agent identity across every surface. Whether a customer uses the website script tag or a WhatsApp message, they interact with the same intelligence. The agent uses the same vector store and the same tool definitions regardless of the communication medium. If you upload a new technical manual to the account-wide file library, that information becomes available to the phone agent and the web agent at the same time. This prevents The Specific Question Trap where leads leave because a generic bot lacks technical depth.

The knowledge for these agents comes from your actual business material. Duvi crawls your website and parses uploaded files into a vector store to ensure answers are grounded in real content. It does not rely on the general knowledge of a base model which might lead to hallucinations. When a customer asks a technical question during a phone call, the agent references the specific documentation you provided rather than guessing. This avoids the issues described in The Contact Us Friction where knowledge remains hidden from the people who need it most.

Answering questions is only the first step. Most support requests eventually require a change in a database or a status update in a CRM. An agent that can only provide text is a message taker. Duvi agents act by calling your own HTTP APIs or MCP servers. When a customer on WhatsApp asks to reschedule an appointment, the agent calls your scheduling API directly. It completes the task instead of telling the customer how to do it themselves. This addresses The Action Gap Crisis where teams are buried in manual tasks that should be automated.

This capability applies to live voice too. The agent handles phone calls using the transcriber and response mode you select in the settings. It can record calls for later analysis and hand over the conversation to a human if the situation requires it. The business connects its own phone numbers or WhatsApp business accounts through providers like Twilio or Meta. This ensures you maintain control over the numbers your customers already recognize.

Launching an agent does not require engineering work to design the conversational logic. You build the agent by talking to Duvi, then provide it with the necessary files and website URLs. Once the agent is ready, you deploy it with a single script tag or by connecting your existing WhatsApp account. Usage runs on a credit system. This allows your support capacity to scale based on conversation volume without the need to manage backend infrastructure.