ZephCraft · decentralized storage

The internet'sshared storage grid.

Every device contributes spare disk into one content-addressed pool — that replicates only what matters. Storage, databases, and compute on one grid. Not the cloud. Not BitTorrent. Not a blockchain.

$ zeph publish ./file see it running →
scroll
live · right now

This isn't a whitepaper. The network is running.

Real numbers, served live by a node on the network. It's early — small on purpose, honest by default. No projections, no vanity metrics.

GETapi.zeph.craft.ec/stats connecting…
nodes online
content stored
coded pieces
provider records
relays
storage used
storage available
total volume contributed
// note A handful of nodes today, growing outward. The point of showing you the real count — even when it's small — is that everything here is built and live, not promised.
the idea

Replicate only what matters.

The internet already has the hardware — billions of devices with idle disk. ZephCraft pools it into one elastic utility. Redundancy is engineered to need, from three knobs:

01 · FLOOR

Erasure coding

4× redundancy survives losing three-quarters of all pieces. Full-copy replication needs ~20× for the same durability. The floor is cheap.

02 · DEMAND

Scaling

Popular content gains providers where there's traffic — and sheds them when demand fades. Temporary, bandwidth-driven.

03 · INTENT

Pinning

Anyone can pin content they care about — a guaranteed anchor. BitTorrent's "seed forever," done deliberately.

— · THE FADE

Forgotten data dies

Content nobody wants and nobody pins is allowed to disappear. Not a bug — the network never pays to keep garbage alive.

redundancy ≈ durability floor + live demand + intentional pins
not like the others

Everyone else's redundancy is an accident, or a bill.

systemhow it replicatesthe catch
BitTorrent Accidental — a side-effect of who's online. Popular = 1000× copies, unpopular = 1× then 0. Wasteful and fragile. Whole-file copies, no erasure, no repair — dies when the last seed leaves.
The cloud Hidden in a data center you rent, at "11 nines" you can't verify. You don't own it. You pay forever. One company can revoke it.
Blockchains Total — every node stores everything. O(N) waste. More nodes ≠ more capacity.
ZephCraft Engineered to need — floor + demand + pins. Deduplicated. Self-healing. Leaner and more durable at once. Distribution, not replication. More nodes = more capacity.
the full vertical · all live

Storage was the floor. The grid runs a stack.

The same network that stores your bytes runs real databases, sandboxed compute, and a durable name registry — every layer built and validated on the live cluster.

L1 · STORAGE

CraftOBJ

Content-addressed objects, self-healing erasure coding, demand-driven lifecycle. Private files are ciphertext-only on the wire.

L2 · DATABASE

CraftSQL

Real SQLite over content-addressed pages. One writer per database, anyone reads — durable and recoverable from the grid even if the owner dies.

L3 · COMPUTE

WASM programs

Fuel-metered sandboxes with capability-scoped SQL + storage. Deploy by name, invoke from any node — the caller's identity comes free with the connection.

L4 · COORDINATION

Head registry

A durable, owner-signed name registry — sharded, writer-elected, resharded live by governance. Your names resolve with you offline. No committee anywhere.

under the hood

A grid, not a backup.

You don't each own a generator sized to your peak — you draw from a grid sized to aggregate demand. Storage should work the same way.

Content addressing · BLAKE3

Every file is named by its hash. Dedup and integrity come for free — the name is the checksum.

Self-healing · RLNC

Any node holding two coded pieces can mint fresh ones — no re-download. Data survives churn no replication scheme can match.

No consensus

Single-writer-per-identity means no blockchain, no gas, no global agreement. It scales linearly, not O(N).

Founder-proof

A keypair is the only credential. Anyone can run a node, a relay, or the routing DHT. If we vanish, the network keeps healing.

join the grid

One binary. Any device. Spare disk becomes the network.

The node runs today on macOS and Linux — connecting across NAT, over its own relays, storing and healing real data. Public builds land with the storage network.

$ zeph # start a node — it finds the network
→ joined · storing pieces · providing to peers
$ zeph publish ./photos.zip
→ pinned · spread across the grid · fetchable by CID, anywhere