Built with
MONA AI, an AI communication platform reachable over WhatsApp and a REST API. Open source.
The premise is that the interface people already use should be the interface. WhatsApp is where a large part of the world's business conversation happens, so the assistant meets users there rather than asking them to adopt another app.
Three models do different jobs. Google Gemini handles text understanding and generation. Gemini Vision interprets images. Stability AI generates and manipulates them. On top of that sit practical capabilities: text and image conversion, optical character recognition, and PDF generation, all triggered from a chat message.
The architectural rule that makes it reliable is strict separation between the listener and the workers. Anything talking to WhatsApp does nothing else, and anything calling a model never blocks message intake. Model calls are slow and occasionally fail, and if that latency reaches the message loop the whole bot appears dead to every user at once, not just the one whose request is running.
Rich media handling covers images, documents, and interactive content, and the REST API exposes the same capabilities to other systems, so the assistant can be a component in a larger workflow rather than only a chat endpoint.
Published open source as a demonstration of AI integration that survives contact with real usage, which mostly means handling rate limits, retries, and partial failure rather than prompt design.
More work