Connecting Support Agents to Live Business Data

A customer asking for a refund status needs a specific update on their transaction instead of a general summary of a return policy. Most AI implementations stop at the knowledge base. They crawl a website or ingest a PDF of FAQs. This creates an agent that can explain a policy but cannot execute a task. It leaves the customer waiting for a human to log into a dashboard. This gap often leads to the specific question trap where users abandon bots that lack depth.

Duvi changes this by allowing agents to call external tools. When you configure an agent, you can grant it access to an HTTP API or a Salesforce instance. You can also use MCP servers. The agent uses these tools to fetch live data or push updates during a conversation. This transition from reading to doing addresses the data write friction that otherwise keeps support workloads high.

Consider a logistics company using a WhatsApp business account. A driver might message to report a vehicle delay. A standard chatbot would provide an office phone number for the driver to call. A Duvi agent can take that WhatsApp message and identify the driver by their phone number. It then triggers an API call to update the shipment status in the internal tracking system. The entire exchange happens in seconds. This prevents the action latency friction where databases remain outdated after a customer inquiry.

The same logic applies to voice calls. If a customer calls your business number, the agent can look up their email address in Salesforce while the call is active. It can see if they have an open ticket about a billing error. The agent can then open the call by confirming it is looking at the existing issue. This removes the need for the customer to repeat their history. Organizations often hire an Agentforce Developer to build these types of CRM integrations. This capability solves the verification deadlock that occurs when agents cannot verify order or serial numbers.

Agent knowledge includes both static files and live data. Duvi crawls your website and parses uploaded files into a vector store to provide context for company policies. The tool calls provide the ground truth for the specific user. By integrating these systems, you cross the tool call threshold and turn conversational bots into functional employees.

Connecting these systems does not require writing a custom integration layer for every channel. You define the tool once in the Duvi Studio. That tool becomes available to the agent for website visitors or phone callers and WhatsApp users. This unified approach eliminates the conversation persistence tax that typically resets the clock on resolution when a user switches devices.

The agent decides when to use a tool based on the conversation. If a user wants to change an appointment, the agent identifies the intent and calls the scheduling API. It collects the necessary parameters, such as the new date, and passes them to your backend. If the API returns a success message, the agent confirms it to the user. If the API fails, the agent can hand the conversation over to a human. The staff member who picks up the chat or call sees the full transcript and the failed tool call. They have the context they need to resolve the case immediately. This structure reduces the cost of the internal ticket transfer.

Deploying an agent this way extends your backend logic directly to your communication channels. Launching an agent requires no engineering work because you describe the agent and Duvi builds it. Usage runs on credits, which keeps costs aligned with the volume of customer problems the agent actually solves.