Architecture

shardd splits the system into two planes.

Private state plane

• full nodes
• local Postgres per full node
• private libp2p mesh
• replication, sync, and convergence

These nodes are correctness-critical and should stay private.

Public edge plane

• regional edge gateways
• public HTTPS
• developer API key auth
• public edge discovery
• full-node selection based on live mesh health

The edge gateway is not a dumb proxy. It participates in the private mesh enough to maintain a live routing view, but it holds no shard state of its own.

Why this is cleaner

• public developers do not need libp2p credentials
• full nodes do not become an internet-facing auth surface
• client tooling stays simple: HTTPS plus a bearer key
• routing decisions use live mesh state, not just DNS

Deployment shape

Typical production layout:

• a few full-node regions
• more public edge regions than full-node regions
• one separate control-plane/dashboard host

Typical dev layout:

• 3 full nodes
• 3 public edges
• 1 dashboard host