About
Why blobs vanish
EIP-4844 gave rollups cheap data availability with a deliberate catch: consensus nodes prune blobs after 4096 epochs, which is roughly 18 days. That is enough for validity windows, and not remotely enough for an L2 node that wants to sync its chain from genesis, an auditor replaying history, or anyone who believes public chain data should stay public. Since Fusaka's PeerDAS, ordinary nodes also custody only a sampled subset of each blob's columns. Complete retrieval is a deliberate, specialized role even inside the window, which is exactly the role this network exists to share.
What this is
A volunteer archive network built on bloar
(Apache-2.0 — blob archive, pronounced to rhyme with "oar"): every blob since Dencun, KZG-verified on ingest, published
over IPFS behind signed, replay-protected head documents. Followers verify
every published claim and pin the authority they select; multiple
independent writers publishing one logical archive make source-schedule
completeness reproducible rather than implicit, so the trust you place in
any single operator, including us, is bounded and checkable.
Who runs it
Built and operated by Tristan Wilson. The protocol is
permissionless: anyone may operate a writer and publish under their own
keys and their own domain. The project does not admit writers to the
protocol and cannot keep them out of it. Human review applies only to the
project-controlled surfaces: *.blobarchive.net DNSLink names
and the shipped default profile catalog. A writer outside those is no
less functional or followable. A faulty writer cannot authorize claims
under another writer's signer or mutate shared CID-addressed bytes, but
it can publish an incomplete history under its own key — followers choose
that trust, and independent derivation exposes disagreement.
What decentralization means here
Two different things get decentralized, by two different roles. Replicas distribute the bytes, the serving capacity, and the failure domains — more copies, more places to fetch from, fewer single points of loss. Independently deriving writers re-derive the history from Ethereum itself and distribute completeness assurance — many copies of one writer's archive do not independently check that writer; a second derivation does. The project controls only its own namespace and default catalog; publishing under the protocol and choosing whom to trust remain open to everyone. The architecture docs map exactly where each property lives.
Contact & support
GitHub Discussions for operator questions. Security-sensitive reports follow the repository security policy rather than public threads. This is a best-effort volunteer service, so there is no SLA.