WhatsApp Knowledge Base Agent
Feed it your business data. Customers ask on WhatsApp — voice, image or text — and it answers in the same format.
The problem
"Customers don't all message the same way. Some send voice notes, some send screenshots of the product, some just type. We couldn't staff replies to all three formats fast enough."
The solution
One flow. Every input type.
A WhatsApp Business webhook fans out on input type. Voice notes are downloaded and transcribed, then handed to the agent as text. Images are downloaded and analyzed by a vision model with the caption as context. Text goes straight to the agent. The agent runs on top of the client's own knowledge base (embedded and retrieved through an MCP client), with tools for web search, calculation and memory. Voice-note senders get an audio reply generated by TTS; everyone else gets a text reply. Same agent, three inputs, matched output format.
Results
Industry
SMB / Customer Support
Service type
AI Automation
Timeline
1 week
Tools used