Most AI chat apps follow the same loop: ask one model, get one answer, hope it’s right. Convergent breaks that. It sends a single prompt to several models at the same time, lays the responses out side by side, and can synthesize them into one combined answer — so you’re validating across models instead of trusting a single one.
It runs as a single HTML file with no framework and no backend. Conversations stay on your device, protected by passphrase-based encryption using the browser’s own crypto APIs, with everything organized into projects and searchable across chats. There’s built-in web search for real-time information, slash commands like /summarize and /explain for faster workflows, and support for just about any provider with an API — OpenAI, Anthropic, Gemini, Groq, Ollama, and more.
Open-source under the MIT license. This one is less about a product and more about a conviction: that a serious tool can be private, local-first, and still just a file you open in a browser — no install, no account, no server watching.