Leaderless coordination for AI agents
Most multi-agent setups need a coordinator: a leader, a queue, a lock server. HiveMind has none. Agents coordinate through the shared hive itself.
Coordination through shared state, not a boss
Each agent reads and writes the same shared memory. Because every agent sees what the others have established, they naturally divide work, hand off, and build on each other without a central authority telling them what to do.
There is no leader to follow, no lock to acquire, and no broker to fall over. Every device is a full, equal participant. Ownership is a separate governance role (who admits devices and sets policy), not a runtime coordinator.
Why leaderless matters
A coordinator is a single point of failure and a bottleneck. Kill it and the system stalls. HiveMind uses a conflict-free merge (a G-Set CRDT over an append-only journal), so any device can keep working offline and the results merge cleanly when it reconnects.
One laptop with its journal is the whole system. Nothing depends on a server being up.
Frequently asked
Does HiveMind use a leader or consensus protocol?
No. Sync has no leader and no Raft or Paxos. Agents coordinate through a shared, append-only memory that merges without conflict, so every device is equal. The hive does have an owner, a governance role that admits devices and sets policy, not a coordinator; if the owner is ever lost, the admitted devices can elect a new one.
What happens if a device goes offline?
It keeps working against its local memory. When it reconnects, its entries merge with the others. No device is special, so nothing stalls.