This lightning talk introduces Suwali, Meedan's multilingual AI platform for trusted knowledge. Today's AI chatbots are trained primarily on English-language content and typically answer with confidence, even when they are wrong. For newsrooms, civil society groups, and human rights organizations, that creates a difficult choice: adopt AI and risk misinforming your audience, or skip in and miss out on the audience growth that AI can help deliver. I'll demonstrate how Suwali takes a different approach. Unlike leading LLM chatbots, Suwali retrieves information directly from a curated multilingual knowledge base built from an organization's own publications, archives, and documents. Every answer is grounded in source material and includes citations. When the answer cannot be found, Suwali simply says, "I don't know." Using a live demonstration, I'll show how this "multilingual library" model enables trustworthy AI that respects local knowledge, works across languages, and helps organizations make their expertise accessible without sacrificing accuracy or control.
