the delivery layerfor open.information.
a 14-gigabyte file posted in berlin reaches a client in tokyo in under a second. the client streams from three peers at once, verifies every chunk with blake3, and pays per megabyte in usdc — whether the payload is a linux iso, an open dataset, a game patch, a scientific archive, or an ai model. deCDN is demand-shaped, locality-optimised delivery for large files at scale: supply forms around demand, cost collapses as regional traffic concentrates. the code is open. the network is open. the price is posted.
open information scaled.its plumbing didn't.
the pattern repeats whenever something big ships: mirrors fork, cdns rate-limit, small teams burn tens of thousands hosting bytes they don't own. deCDN inverts every axis — supply forms around demand, not allocated to it.
How it works
a single quic 0-rtt handshake broadcasts the blake3 hash to a kademlia dht of nearby peers. each node answers with what it has cached, its current rate per megabyte, and how fast it can serve — p50 under 100 milliseconds. the client ranks the answers by price, latency, and reputation; the best-priced, fastest, most-reputable node wins the request, or several win in parallel for a large file.
bytes flow over quic directly from the chosen node; for files over ten gigabytes the client opens parallel streams to several peers at once and aggregates their throughput — a 1 gbps origin turns into multi-gigabit delivery to the client. every chunk is verified against the blake3 tree hash the instant it lands; tampered bytes trigger immediate disconnect and a fraud proof against the node's stake. trust no node — verify every byte.
you pay per megabyte in usdc, automatically, as the bytes arrive — no monthly invoice, no subscription, no whole-file minimum. pay for what you pulled, nothing more.
frequently asked.
Legacy CDNs provision for peak traffic and amortise that cost across long contracts, so a popular release gets expensive fast. deCDN is demand-shaped: supply forms around traffic, nodes cache what's hot, and bandwidth gets reused across many pulls. Cost per gigabyte collapses as regional demand concentrates — the opposite of how fixed-provisioning CDNs behave. Target price today: ~$0.01/GB, one cent, per actual gigabyte delivered.
Every blob is BLAKE3-addressed, so the hash the client is asking for is the same hash they'll be checking on the way in. Verification happens chunk-by-chunk, in flight — mismatched bytes are dropped and the next peer on the list is asked instead. Peers that send garbage lose staked TOKEN via a non-custodial Merkle proof. There is no central arbiter; the math is the arbiter.
TOKEN secures the network — operators stake it to participate, and they lose it if they misbehave. Payments, though, happen in USDC, because operators want stable currency they can pay power bills with. Twenty percent of protocol fees flow into a Balancer V3 80/20 TOKEN/USDC buyback, tying token demand to network throughput.
The client pulling the bytes. When a session starts, the client opens an off-chain USDC channel and streams payment per megabyte as chunks arrive. No subscription, no account, no invoice — when the download finishes, the channel closes and settles. You pay for what you pulled, nothing more.
The client detects the BLAKE3 mismatch on the very next chunk, drops the connection, and files a non-custodial Merkle proof against the node's stake. The good node further down the list keeps streaming — the client barely notices. The bad node loses TOKEN on-chain within minutes. There is no central arbiter; the math is the arbiter.
Yes — AI is where deCDN's primitives compound. Clients pull large weights from several peers in parallel and saturate their own uplink instead of a single origin's. BLAKE3 chunk addressing lets teams grab one quantisation variant or a single layer without downloading the full checkpoint. And when a model drops and every region pulls it at once, the mesh warms around the demand instead of choking at a central bucket.
Contact
deCDNis a peer-to-peer delivery layer for large files — open datasets, scientific archives, linux distros, game builds, video libraries, open-weight ai models, anything meant to be pulled a million times and that shouldn't depend on one bucket. the code is open. the network is open. the price is posted.