Yash
    05 / 05·Beta·All case studies

    CodeFusionAI

    Collaborative coding with isolated execution and realtime sync.

    GitHub
    CodeFusionAI — project preview

    Overview

    CodeFusionAI is a collaborative coding environment combining shared editing workflows, realtime synchronization, isolated execution containers, and multi-user presence systems.

    The project explores how collaborative editing and shared execution can coexist without tightly coupling transport, presence, and runtime systems.

    Problem

    Collaborative editors become difficult to maintain once synchronization, execution isolation, and realtime presence all interact simultaneously.

    Many systems either overcomplicate operational transforms or tightly couple cursor traffic with document synchronization.

    The goal was to build a simpler architecture focused on correctness and predictable behavior.

    Architecture

    Backend systems are written in Go with isolated Docker execution environments and cgroup-limited runtime sandboxes.

    Realtime coordination uses WebSockets with separate transport channels for document synchronization and presence traffic.

    Frontend systems are built with React and shared multi-user editor workflows.

    Document

    1. EditorReact
    2. OT layerGo; small and well tested
    3. Shared documentOne converged state

    Presence

    1. Cursors and selectionsHigh-frequency, disposable
    2. Separate WebSocket channelNever queued behind edits

    Execution

    1. Run requestFrom the session
    2. Docker containerOne per session
    3. cgroup limitsCPU and memory bounded
    Edits, presence and execution each get their own path, so none can stall the others.

    Tradeoffs

    The project intentionally favors operational correctness over aggressive feature expansion.

    Certain IDE-style capabilities and advanced collaboration layers were intentionally not built to keep synchronization behavior understandable and maintainable.

    Sandboxed
    Execution
    OT layer
    Sync
    Split
    Channels

    What I'd do differently

    The project improved understanding of OT systems, synchronization boundaries, isolated execution environments, realtime transport separation, and collaborative runtime coordination.

    It also reinforced how quickly complexity grows when shared execution and shared editing interact.