SEDRAIntroduction
For most of its history, the blockchain served one relationship: a person sending value to another person. The relationship that will define the next decade is different. It is software acting for itself.
The rise of autonomous software
Agents built on modern models are moving from answering questions to taking actions. They book, buy, write, schedule, and negotiate, and increasingly they do so without a human waiting on each step. As they grow more capable they will transact constantly, on behalf of people and companies, and more and more with one another. This is not a distant scenario. It is the direction the whole software industry is already moving.
Why today's infrastructure falls short
When two agents from different makers need to agree on what happened, settle a payment, and carry on, neither can be asked to trust the other's private servers. They need common ground that belongs to no one in particular. The chains built for human payments were not designed for machine-speed concurrency, and they slow down exactly when many actors move at once. The platforms built by the large model labs are private by design, and a private platform cannot be neutral ground for the whole field.
What Sedra is
Sedra is a proof-of-work layer-one built on GHOSTDAG parallel consensus, with kHeavyHash proof of work and a one-second target block time. It launched in 2024, went dark, and has been restored in the open, live again since August 2026 with public mining open to everyone. Its purpose is to be the neutral place where autonomous agents coordinate, remember, prove, and pay. The rest of this paper explains how.
The parallel insight
The whole network rests on one idea. Intelligence is not a single line of thought. It is many things happening at once. Agents do not wait their turn; they act together, in the same instant, by the thousands.
The ceiling of serial consensus
Almost every blockchain elects one leader for each slot, accepts one block, and discards whatever else was produced at that moment. Everything is forced into a single file. Under a crowd of concurrent actors this design does not merely slow down. It works against the very thing it is meant to serve, and the harder the network is used, the more it fights itself.
GHOSTDAG and PHANTOM
Sedra uses GHOSTDAG, a block-DAG protocol from the PHANTOM line of research. Instead of a single chain of blocks, blocks reference many predecessors and form a directed acyclic graph. The protocol identifies a well-connected cluster of honest blocks, then derives a single total order across the whole graph, keeping the blocks produced in parallel rather than throwing them away. Security holds under the same honest-majority assumption as a longest-chain design, while throughput is no longer bottlenecked by a one-block-at-a-time race.
Why this is the right shape
Concurrency is not an edge case Sedra tolerates. It is the normal case it was designed for. When many agents write in the same instant, the network does not ask them to queue; it accepts the parallel reality and settles it into one order everyone shares. The structure of the ledger matches the structure of intelligence, and that single alignment is what makes everything else in this paper possible.
Protocol architecture
Sedra keeps the base chain simple and fast, and pushes heavy computation to the edges where the hardware lives. The result is a settlement and record layer that stays lean while the ambitious work happens around it.
| Consensus | GHOSTDAG with the PHANTOM ordering rule, producing one canonical order over parallel blocks. |
| Proof of work | kHeavyHash, shared with the wider GHOSTDAG family, so existing mining hardware works with nothing new to buy. |
| Block time | A one-second target, for confirmation that keeps pace with machine activity. |
| Ledger model | UTXO, precise and auditable, well suited to settlement and to anchoring records. |
| Reference node | sedrad, written in Go, open source, with a standard RPC and peer-to-peer interface. |
Ordering and confirmation
Because blocks arrive in parallel and are ordered together, confirmation is fast and grows more certain as the graph deepens over a block. A one-second target keeps the flow steady, which matters when the participants are machines that do not pause between actions.
Off-chain compute, on-chain settlement
Sedra does not try to run heavy computation on the base layer. Models run, and work is done, on the machines that own the hardware. The chain settles value and records what happened. New capabilities are built at the application layer first, where they can be reviewed and revised quickly, and only hardened into consensus once they are proven. This lets the network move quickly without ever putting the chain at risk, and it keeps consensus rules fully deterministic: analysis and tooling may recommend, but nothing adjusts live consensus based on model inference.
kHeavyHash and the shape of AI compute
The proof of work is worth a closer look, because its shape is not an accident. At its core, kHeavyHash is a matrix multiplication wrapped between two standard hashes. Matrix multiplication is also the operation that runs neural networks, so the work that secures Sedra is the same kind of work that runs AI. The algorithm descends from Optical Proof of Work, a design meant for silicon-photonic hardware that performs these multiplications with light, moving the cost of mining out of electricity and into the chip itself.
That same photonic hardware is one of the most promising answers to AI's own energy problem, as data-center power demand races toward roughly double its 2025 level by the end of the decade and AI runs into the limits of the grid. As those chips arrive, the hardware that makes AI efficient and the hardware that secures Sedra converge. Proof of work stops competing with useful compute for the same electricity and starts pointing at the same future.
Optical mining is still experimental. Today kHeavyHash runs on ordinary GPUs, FPGAs, and ASICs, and mining rigs do not run AI models. The alignment described here is real but forward-looking: a shared core operation, a shared hardware direction, and an energy model that rewards efficient chips. We mark it as direction, not a claim about today.
The agent primitives
Everything an autonomous agent needs from neutral ground comes down to four capabilities, each playing to what a fast, ordered, honest ledger does best.
A shared memory
Countless agents write to one ordered record at once, and Sedra keeps all of it. They work from a single history they can trust, instead of fighting over whose version is right. This is Consensus Memory, and it maps directly onto GHOSTDAG's ability to merge concurrent writes into one agreed sequence.
A cost to mean it
Every action carries a small, real cost. Spam and imposters fall away, and a commitment finally stands for something because making it was not free. Proof of intent turns a costless message into a credible signal.
A name they carry
An agent builds a verifiable track record, held by the network rather than any one company or model. A reputation it carries anywhere, that cannot be faked or taken away. Of all the primitives this is the stickiest, because a longitudinal, attested history cannot be moved or forged.
A record of what is real
As machines flood the world with content, Sedra keeps an honest, time-stamped account of what was said and done, so anyone can check what actually happened. It is a notary for machine work, cheap enough to use at scale.
These are the direction Sedra is built toward. The ordering, settlement, identity, and record-keeping beneath them are what a proof-of-work GHOSTDAG chain provides today. The higher-level agent features are developed in the open, application-first, and marked plainly as research until they are live.
Decentralized intelligence
The concentration problem
Today a handful of companies hold the models, the compute, and the memory behind almost every machine that thinks. That is a concentration of power over intelligence itself, and it grows tighter as models get larger and more expensive to run. A neutral coordination layer is only half the answer if the intelligence it coordinates lives inside a few private clouds.
The worker network
Sedra points the other way. Anyone with a capable machine can join as a worker, running any model, across any kind of task, from text and images to sound, video, and actions in the world. Sedra does not run the models; the computing stays with the people who own the hardware. What the network provides is the connective tissue a scattered crowd needs to act as one: an identity to be known by, a shared memory to write to, proof of the work performed, and payment that settles on its own.
Parallel work, one result
A large job splits across many workers at once, and their parallel results are ordered the same way Sedra orders everything else. The same property that makes the chain suited to many agents makes it suited to many workers. It is the ground the models grow from, not the models themselves.
Trust in the results
Verifying that a worker did its job honestly is a genuine research problem, and this paper does not pretend otherwise. The honest path is to start with mechanisms that are recoverable if they are imperfect: reputation earned over time, redundant execution with agreement between independent workers, and a refundable bond that can be forfeited for bad work. Stronger guarantees, such as hardware attestation and zero-knowledge proofs of computation, are a direction to grow into once the simpler approach has run and been reviewed. Depth of guarantee should follow proof, not precede it.
Metacore, the first layer live
Metacore is the first working piece of this: a network for verifiable AI work, live today at app.sedracoin.com. A task is handled by several independent models, graded by independent judges, and returned as the best answer with a portable receipt sealed to Sedra that anyone can check. It is paid in SDR, refunded if it cannot be verified, and the largest share of each fee goes to the workers who also mine, giving proof of work its demand side directly. Because every answer is proven and portable, answers can relate to one another: new work can cite verified answers as trusted inputs, cross-check them, and challenge or supersede them. This turns one-off AI output into a compounding, auditable web of knowledge, something a private, unverifiable model cannot offer. The open-ended answering path is rolling out as the judge network comes online, and we mark plainly what is live versus in progress.
Token and economics
SDR and its supply
SDR is the native asset. It is earned through proof-of-work mining under a fixed maximum cap, and the genesis block carried no premine. The code, the supply, and the rules that govern them are public and verifiable on-chain. The cap is a ceiling, and it is not one we raise.
What SDR is for
SDR plays two roles. It is the reward that pays for the network's security, aligning miners with the health of the chain. And it is the fuel agents spend to write to shared memory, to signal intent, and to pay one another. Demand for SDR is meant to come from real usage rather than speculation, and to grow with the activity on the network.
One honest money
SDR is one honest money, deliberately one asset rather than a pair of tokens. On Sedra it is earned through proof of work, and it settles, remembers, proves, and pays: it is the fuel agents spend, the slashable, refundable bond they post on their work, and the reward that secures the chain. Everything the economy needs runs here, in SDR. Richer deal mechanics, escrow that pays on delivery, streaming payments as work is used, and royalties that keep paying whoever made something that gets reused, are built application-first and settle on the base. Carrying that programmability and reach to other chains through a wrapped SDR, backed one to one by reserves anyone can check, is later roadmap work. The base stays simple and honest.
Allocation policy
Where any coded allocation is ever used, it follows the integrity principles in the next section without exception: written into the code rather than mined by hand, its full amount and address disclosed with the release, held by many keys in a multisig rather than one, and released on a vesting schedule and timelock that anyone can watch, with every outflow logged against its stated purpose.
Integrity
Trust on Sedra does not rest on anyone's word. It is written into the code and open to inspection. Sedra was rebuilt from the ground up, in the open, with the rules below binding from the first block of its new life.
Integrity principles
- No off-schedule supply. Every change ships as a public, versioned release, never a quietly altered client.
- No head start. Mining opens to everyone at the same announced moment, the team included.
- Any allocation is written into the code, not mined by hand.
- Its full amount and address are disclosed with the release, not after.
- It is held by many keys in a multisig, never a single one.
- It is released on a vesting schedule and timelock that anyone can watch.
Keeping the record honest
- Integrity Ledger. The project's history and these principles anchored on-chain, so the record can never quietly drift from the truth.
- External anchoring. A periodic hash of Sedra's chain state stamped into an independent chain Sedra does not control, so the past cannot be rewritten unnoticed.
- Proof-of-Spend. Treasury outflows verifiable against their stated purpose, surfaced through the wallet's steward view.
Governance
Anonymous does not have to mean unaccountable. Sedra is governed through three functional roles tied to consistent cryptographic identities over time, not named individuals.
- Technical steward. Protocol and code decisions, and release approval.
- Treasury steward. Allocation, spending, and vesting, exercised only through the shared multisig, never unilaterally.
- Community steward. Growth, engagement, and public communication.
Any decision above a defined threshold requires more than one steward to sign, the same multisig mechanism that governs the treasury applied to governance itself. Accountability comes from consistency and shared keys, not from unmasking anyone. Over time, more of these decisions open to the wider community of operators who carry the network.
Durability and decentralization
A network meant to be neutral ground cannot depend on the people who started it. Sedra once went dark because it leaned on too few servers. The guiding rule now is simple: the network must keep running even if any single operator disappears, ourselves included, at every layer.
Discovery and seeds
Multiple independent seed domains, a DNS seeder, and hardcoded fallback peers mean that joining the network never rests on one address or one operator staying online.
Nodes and operators
Independent operators run full nodes alongside the team, with a one-choice option built into the desktop and command-line apps so that helping the network takes a single click rather than a manual setup.
Mining and liveness
Solo miners and reconnected pools produce blocks and provide real liveness. Because Sedra shares kHeavyHash with the wider GHOSTDAG family, existing rigs work from day one, and a fresh restart means difficulty opens low, a genuine and fair opportunity for returning miners. Those rigs do more than mint the coin: their proof of work secures the autonomous work that runs on Sedra, so mining here stands behind real machine activity, not payments alone.
Mining works on two layers, and pools are first-class. Hashpower, whether solo or pooled, secures the chain, and because the chain is the shared memory, hashpower secures that memory; a pooled miner changes nothing to take part. Separately, node operators, including pools, can run the memory and coordination roles as they arrive on the roadmap: storing and serving the on-chain memory, and coordinating and verifying the agent economy. Securing is paid by block rewards; those node roles are paid by the fees the agent economy pays, not by new issuance. To stay decentralized, both layers must stay spread: hashrate across many pools and solo miners, and the coordination role permissionless across many operators, with verification decided by a quorum so no single operator or pool decides what is true.
Backups and checkpoints
A disciplined offsite backup practice and a periodically published checkpoint keep the chain recoverable in any worst case, so the near-loss that once threatened the project can never repeat.
The attestation agent
An open-source, independently reproducible agent verifies which operators are genuinely online and synced. Its own determinism matters as much as consensus, and it is cross-checked across instances before it is trusted to drive any reward.
The participant experience
One wallet, three tiers, with an explorer built to the same standard. Not a scatter of disconnected tools, but one coherent experience.
One wallet, three tiers
- Basic. Non-custodial send, receive, and balance, with keys generated on the user's own device.
- Contributor. On desktop and command line, a plain choice to also help run the network, which earns a disclosed reward for verified uptime.
- Steward. Not a login but a wallet that gets opened. Multisig co-signing, vesting visibility, and spend-logging appear only when the wallet holds a key to the treasury address itself, so the allocation is what unlocks the tools, not a permission flag layered on top.
Protection, built in
- Address Sentinel. Defenses against address-poisoning at every tier: dust transfers never populate the recent list, near-match addresses are flagged before reuse, and large sends demand explicit confirmation. Address poisoning has cost the wider industry heavily, and these defenses are on by default.
- Continuity Vault. Recovery split across trusted parties so no one, including the team, holds a complete key, paired with an optional inheritance path that activates only after a long period of inactivity and stays cancellable while the owner is active.
- An explorer that explains. Plain-language summaries of transactions and blocks, natural search over chain data, and clear labeling of known entity types, built for understanding rather than raw hex.
Earning and utility
Two honest mechanisms, kept deliberately separate because they mean different things and carry different risk.
- Contribution reward. A small, disclosed reward for verified infrastructure contribution, such as keeping a node online. It is not staking and carries no slashing, and it touches nothing in proof-of-work security. It ships as application logic that can be reviewed and revised, and earns deeper protocol integration only after it has proven itself. It is described plainly against its honest risks rather than as a one-sided yield pitch.
- Locked utility. SDR can be voluntarily time-locked, on-chain, to unlock service benefits such as expanded access or priority, rather than a financial return. It is utility for locking, kept clearly distinct from any notion of investment yield, in both disclosure and code.
The ecosystem
The pieces that turn the idea into something builders can hold. Each is developed in the open, with research and shipped code kept clearly apart.
- Metacore, the agent kit, and the worker client. Metacore is live now: verifiable AI work anyone can use at app.sedracoin.com, with the desktop app letting any capable machine join as a worker, run a local model, and be paid. Around it, a small open kit so any agent can plug into Sedra in a few lines, for a wallet, a shared memory, a name, and payment, meeting developers inside the tools they already use.
- Proof-of-work for agents. A request path, AgentPort, where an agent attaches a small, genuine proof of work as a signal of seriousness, and builds reputation from its record of commitments kept versus broken.
- wSDR. A one-to-one wrapped SDR that carries the asset to smart-contract chains and open liquidity, backed by reserves anyone can check, while the base chain stays honest proof of work. It reaches the markets and the agents where they already are.
- The GHOSTDAG bridge. Interoperability to move value between Sedra and the wider GHOSTDAG family, so neighboring networks strengthen each other rather than wall themselves off.
Roadmap
Six steps, in order, each built on the one before it and honest about where it stands.
- The network is back. Done. The chain is restored and the nodes, explorer, and wallet are online, in the open.
- A fair restart. Live. Public mining is open to everyone, live since August 2026, with the full supply and treasury address public. The same rigs that mine the wider GHOSTDAG family work here from day one.
- Tools for everyone. Live. A clean wallet and explorer for everyday use, on desktop and phone, plus a desktop app to run a node, mine, and earn in a single click.
- Metacore. Live, and deepening. Verifiable AI in use today: independent models, independent judges, and answers sealed to Sedra that relate to one another. The open-ended judge network and the linked-proof relations are rolling out now.
- Built to last. Many independent operators, shared-key governance, and public reporting on treasury, runway, and network health. It runs even if any one of us steps away.
- The age of agents. Shared memory for agents, an open kit so any agent or worker can plug in, ownership of what they build, and wider liquidity across chains.
Getting involved
Sedra is being laid down in the open, and it is early. Run a node and help carry the network. Point a miner at it, mining is open to everyone now. Put Metacore to work today, or build the first agents and workers that will live here. Or simply watch in the open and hold us to the principles. This is the ground floor of the economy that comes next, and it is being built to belong to the people who show up for it.
This paper describes a network under active development. It is information, not investment solicitation or financial advice. SDR figures are real where stated and are confirmed against live chain state, and anything not yet live is described as work still ahead. Nothing here is a promise of financial return.