Distributed storage that heals itself

curl -sSfL https://sh.getnauka.com | sh

Nauka splits every file into Reed-Solomon shards scattered across the nodes of a cluster. As long as k shards per stripe survive somewhere, the file comes back byte-for-byte identical.

Membership is explicit and yours: the first machine founds the cluster, every next one is added with a single command that provisions it over SSH and takes it through consensus. No discovery layer, no central server, no side infrastructure to keep alive.

# first machine — installs nauka and founds a systemd-managed cluster
curl -sSfL https://sh.getnauka.com | sh

# every next machine, run from the first one
nauka node add <ip>:7311

Durable by arithmetic, not by copying

4+2 by default: every stripe survives losing any 2 shards out of 6, for a 50% storage overhead — where ×3 replication charges 200% for the same tolerance. BLAKE3 integrity is checked at every boundary; a corrupted shard is caught on read and treated as lost, never served.

It repairs itself

A scrubber checks every file continuously and regenerates missing shards from the survivors. A node that dies is healed around; a node that comes back is refilled; a node you remove is drained. Nothing to babysit — what survives what is spelled out, not implied.

Built for machines that are far apart

Nodes learn their network positions from the round-trip times they measure (no GeoIP database) and pull each stripe’s shards apart, so a file survives the loss of a region — not merely of a machine. Metered links get monthly egress budgets: exhausted nodes are deprioritized, never refused.

Proofs, not claims

A node can assert it still holds what it has quietly lost. Nauka demands blake3(nonce ‖ bytes) proofs of possession before it gives up any redundancy, and audits its peers continuously by sampling.

How it compares

	Nauka	Garage	MinIO	IPFS
Erasure coding (no ×3 replication)	✅	❌	✅	❌
Self-healing	✅	partial	✅	❌
Single binary	✅	✅	~	~
Capacity-weighted placement	✅	✅	❌	❌
Topology-aware placement (learned, not configured)	✅	❌	❌	❌

Nauka is the storage engine, deliberately bare: it exposes an HTTP API and — behind an opt-in build flag — an S3-compatible endpoint. User-facing applications belong in products built on top, not in the engine.

Where to go

• Quickstart — a real cluster in five minutes.
• Durability & consistency — the contract: what survives what, and how we know.
• Deploy — keys, ports, systemd, sizing; the long version.
• CLI reference — every command, with examples that were run for real.