Decentralized Spatial Compute Mesh & Multi-Channel Routing.

1. Why a cloud server slows collaborative coding down

Most collaborative editors send every keystroke, compiler message and video stream through a central cloud server. Pair-debugging between Tokyo and San Francisco that way adds 140 to 280 ms of round-trip delay, because each character has to be acknowledged by the server before your teammate sees it.

It gets worse with AI agents. When Claude Code, Gemini or @CruxAI edit hundreds of lines at once, the server becomes a queue: the editor freezes, diffs merge out of order and code breaks.

In CruxYour edits go straight to your teammate, so pairing across continents feels like sitting side by side.

2. The 7×7 core and its four channels

Crux models a session like a chip. The center 7×7 grid (49 tiles) is the shared workspace. Four chips inside it each carry one kind of data on its own channel:

• Text: code edits as syntax-tree changes over encrypted WebRTC data channels, converging in under 10 ms.
• Voice: low-latency Opus audio, with no Zoom or Discord running in the background.
• Video: hardware-encoded H.264 camera feeds shown inside the editor.
• Screens: a shared GPU canvas with everyone's live pointer.

In CruxCode, voice, video and screens each travel separately, so a heavy video call never slows down your typing.

3. Staying in sync without a coordinator

With no central server, Crux needs another way to agree on the order of edits. Every change is a signed node in a Merkle DAG, ordered with a Lamport clock:

// Merkle node for decentralized syntax convergence
pub struct MerkleNode {
    pub hash: [u8; 32],
    pub parent_hashes: Vec<[u8; 32]>,
    pub lamport_clock: u64,
    pub peer_id: PeerId,
    pub payload: SyntaxDeltaPayload,
}

impl MerkleNode {
    pub fn verify_signature(&self) -> bool {
        // Curve25519 signature verification on-device
        crypto::verify(&self.hash, &self.peer_id.pubkey)
    }
}

When a peer drops off and comes back (say, you close your laptop on a train), Crux compares Merkle roots to find what changed in O(log N) time and merges it with no lost keystrokes.

In CruxClose your laptop mid-edit. When you're back online, everything merges and nothing is lost.