Is CoCalc Possible?
CoCalc: Is it Possible to Build a Good Collaborative Scientific Computation Platform? https://cocalc.com/wstein/omnibase/2025-12-05-umn CoCalc: Is it Possible to Build a Good Collaborative Scientific Computation Platform? William Stein (SageMath Inc.) at the UMN Industrial Problems Seminar, Dec 5, 2025 In this talk, William Stein traces his path from number theory and founding SageMath to building CoCalc—a real-time, collaborative platform for scientific computing and teaching. He explains why CoCalc was created (research clusters, teaching at scale, early browser notebooks), early business and IP hurdles with UW resolved via NSF open-source requirements, and the shift from academia to a product company. Stein details how CoCalc differs from SageMath (realtime collab, course management, evolving web stack, tighter licensing), the platform’s abuse and compliance realities, and demos features like shared terminals, Jupyter, LaTeX, time-travel history, snapshots, chat, whiteboards/slides, and AI. He candidly reviews today’s challenges (bureaucracy, competition like free Colab EDU, performance, AI agent integration, easy install) and outlines a deep architectural rewrite: no networked filesystem, browser-to-project websocket connections on nearby servers, horizontal scaling with pub/sub, and a small self-contained binary that runs locally or via SSH—with optional sync and AI. He conjectures a “yes”: a good platform is possible, but requires rethinking the foundations. Learn more: https://cocalc.com and talk resources: https://cocalc.com/wstein/omnibase/2025-12-05-umn Key topics: Origins: SageMath, research clusters, early web notebooks, teaching at scale Business/IP: UW claims, NSF open-source loophole, leaving academia Product reality: collab features, course management, compliance, abuse handling Competition and priorities: performance, AI agent integration, simple installs New architecture: direct browser→server websockets, regional servers, custom pub/sub, container-free binary Outlook: CoCalc can work—by modernizing core assumptions and execution