I built a WhatsApp shop assistant that runs on a CPU-only laptop. The hard part was getting it to say "I don't know."

Developer Newton Lebniz has created SellMate, an open-source AI shop assistant designed to help solo sellers manage product listings and customer inquiries on WhatsApp. The tool runs entirely locally on a standard laptop without a dedicated GPU, utilizing Ollama to host Moondream for image analysis and Gemma 3 1B for text generation. A key feature of the project is its "claim guard," which prevents the AI from hallucinating product details by comparing generated descriptions against seller-provided facts. When the model cannot answer a customer's question based on the provided data, it is designed to transparently admit ignorance and route the inquiry to the seller's inbox. This project highlights the potential for running efficient, privacy-focused AI pipelines on consumer hardware, emphasizing that process-driven guardrails are often more critical for reliability than model size alone. The project is available on GitHub for those interested in local, low-cost automation.