The infrastructure layer for permanent data and decentralized intelligent services.
Cosmos SDK · IBC
Wallet support
Cosmos chain · IBC
Cosmos chain · IBC
Ecosystem campaigns
Institutional validator
Market data
ZK Layer 1
Cosmos chain · IBC
Cosmos SDK · IBC
Wallet support
Cosmos chain · IBC
Cosmos chain · IBC
Ecosystem campaigns
Institutional validator
Market data
ZK Layer 1
Cosmos chain · IBC
Modern applications increasingly rely on permanent data, identity, execution and verification, and today each one comes from a different vendor. Lumera brings them together on one architecture. The more autonomous the software gets — agents that act, spend and decide without a person in the loop — the more it needs all four at once.
Every capability is built on the same architecture and execution framework, so Lumera grows through composition rather than as separate systems.
Explore the architecture →Pay once and the network keeps your data for good.
import { Cascade } from "@lumera-protocol/sdk-js"
const cascade = new Cascade(client)
const result = await cascade.upload(file)
console.log(result.txHash)
console.log(result.contentId)
import { Lumera } from "@lumera-protocol/sdk-js"
const lumera = new Lumera({ apiKey: process.env.LUMERA_API_KEY })
const memory = await lumera.memory.open(agentId)
await memory.remember({
user: "Alice",
preference: "Window seat"
})
const context = await memory.recall()
The same primitive in two shapes: a file an application stores, or a fact an agent needs to still have next session.
Lumera separates consensus from decentralized services. The blockchain records requests, payments, governance, and cryptographic proofs while economically bonded SuperNodes perform storage, verification, and decentralized execution. This allows sophisticated protocol services to scale independently without compromising security or decentralization.
Write to the network, hold data for it, or secure it. Each one is open today.
Pick the one that matches what you need.
Permanent data, identity, verifiable execution, settlement and trust on a single protocol, with memory live on mainnet and each capability built on the one beneath it.
Durable memory is the layer that shipped first, because none of the others work without it. Hover any capability to see what it rests on and which parts of the network it uses.
The gaps are architectural, not incremental, and they depend on each other. Identity without settlement is unenforceable. Settlement without identity is unaccountable. Open any row to see what breaks today and what replaces it.
An agent’s identity is an API key belonging to someone else. Revocable, non-portable, carrying no independent reputation. It cannot accumulate a record, delegate authority, or be held accountable for anything.
Lumera ID is the universal on-chain principal for users, agents, applications, data assets and providers. It is a container that accumulates claims over time.
Delegation chains are the enterprise wedge. Grant an agent a budget class or a dataset. Scoped, time-bound, revocable, enforced on-chain.
Nothing proves what model ran, what data it consumed, what it cost, or whether it was the model advertised. The client trusts the provider completely.
Agent state lives in provider-controlled databases. Switch providers, lose context. No memory the agent owns, no institutional knowledge that accumulates, no data sovereignty.
Cascade is content-addressed erasure-coded storage with BLAKE3 Merkle proofs. Alongside it sits a leaderless quorum-replicated database for hot state, the tier an application reads from between writes. Large values overflow from one to the other transparently.
Payment rails assume a human at the end. An agent cannot hold a balance, earn revenue, or settle a transaction on its own behalf.
LUME is the settlement, staking and governance token. Everlight funds what per-use fees cannot: an artifact that must stay available for years, a verifier ready before it is needed, a node paid for capacity.
See the token model →Three classes of execution, three consequences when something goes wrong. They never collapse into each other.
Quality moves the trust graph. Nothing is slashed.
If runner and verifier disagree, that is objective evidence. No judgement needed.
The protocol never fires an irreversible action on its own.
LUME is the settlement asset, the security bond, the reputation stake and the governance weight. Each job is tied to activity you can look up on the explorer.
Every storage write, retention payment, verification job and agent session settles in LUME. Demand scales with real economic activity, not with speculation about it.
SuperNodes bond LUME to join the execution plane. Bonds are slashable for objective misbehavior: a deterministic mismatch, a forged receipt, a non-response. More capital flowing through the network requires proportionally more locked LUME.
Staking turns reputation from a score into a commitment that can be lost. An operator or agent with staked reputation has something real at risk.
Parameter changes, fee adjustments and upgrades require LUME-weighted votes, amplified by standing in the trust graph, so those who have done the work carry more weight.
Each turn of the wheel makes the next one cheaper to win. That is the whole argument.
Permanent storage, identity, verifiable execution and settlement behind one SDK and one economic model. Build against stable interfaces.
Fund a Global Credits balance once, on-chain. Provider-agnostic from the start.
Written once, readable from any framework, agent, or model, and still provably there decades from now.
An agent that forgets is a demo. Every session starts from nothing, the same context gets re-sent and re-paid for, and anything the agent learned about a customer last week is gone.
The memory that does exist usually sits in a database on someone else’s account. It can be deleted, it can be lost in a migration, and there is no way to show a third party that a record has not been quietly edited.
Lumera stores each fact once, ties it to an identity the agent owns rather than to a vendor account, and keeps re-proving that the data is still held. Your agent reads and writes through an SDK. There is no wallet, no gas and no chain to learn.
Pay once and the fact is kept. There is no renewal to forget and no bill that grows with age.
The network keeps re-proving that the data is still held, continuously, and anyone can check the result.
Memory belongs to the identity that wrote it, not to the application or vendor it was written through. Switch model, switch framework, keep the memory.
Paying an operator to hold data is not the same as knowing they still have it. The network settles that question continuously, on its own, without anyone asking.
import { Lumera } from '@lumera-protocol/sdk-js'
const lumera = new Lumera({ apiKey: process.env.LUMERA_API_KEY })
// 1. open the memory an agent writes under
const memory = await lumera.memory.open(agentId)
// 2. remember a fact once, it is kept forever
await memory.remember({ units: 'metric' })
// 3. recall it from any framework or model
const context = await memory.recall()
// 4. ask the network to prove it is still there
const proof = await memory.prove(context.id)
No. A vector database makes memory searchable and object storage makes it cheap; neither can prove the data is still there, and both tie the memory to an account you rent. Lumera sits underneath as the permanent, provable copy the agent owns: pay once instead of monthly, retention proven by Merkle challenge instead of promised in an SLA, and every read and write adding to the record the agent carries to its next counterparty. Keep your index, point it here.
No. You call the SDK with an API key, the same as any other service. That key is how you authenticate to Lumera. It is separate from the on-chain identity your data is written under. There is no gas to budget, no token to hold, and no seed phrase for your users.
Nothing is lost. You pay once at the point of writing, and operators keep being compensated for holding it after that, continuously, out of protocol rewards and a share of network fees. Retention does not depend on a subscription staying current, which is the difference between permanence funded and permanence promised.
Yes, and you do not swap anything out. The MCP connector mounts underneath the harness you already run, so an agent in LangChain, CrewAI or anything else that speaks MCP keeps its loop exactly as it is and gains memory beneath it. There is also a Python and TypeScript SDK, and an HTTP API for everything else.
Not in the sense of erasing it, and that is deliberate: a memory you could quietly remove is not evidence. What you control is what goes in, and you can supersede a record so the current answer changes while the history stays intact. Treat it the way you would an append-only ledger, and keep anything you may need to erase out of it.
Three calls and an API key. The network keeps the data and keeps proving it is still there.
Cryptographic passports that enforce permissions and delegated authority for people, applications and autonomous systems, so an action can always be traced to whoever was allowed to take it.
An API key is a secret that grants access. An identity is a thing that accumulates: what it may do, what it has done, and who it lent authority to.
The developer documentation carries the detail this page leaves out: the interfaces, the guarantees, and how it fits with everything else.
Actions that produce cryptographic receipts proving what work was performed, so a result can be re-derived by someone else rather than taken on faith.
A receipt is not a log line saying the job finished. It carries everything a second party needs to do the work themselves and compare.
The developer documentation carries the detail this page leaves out: the interfaces, the guarantees, and how it fits with everything else.
Protocol-native settlement, escrow and incentives for decentralized services, so an application or an agent can hold a budget, pay a provider, and be paid itself.
There is no invoice, no net-30, and no dispute window where the money sits with whoever happens to hold it. Each stage is a state the protocol enforces.
The developer documentation carries the detail this page leaves out: the interfaces, the guarantees, and how it fits with everything else.
Trust built from verifiable protocol activity rather than from intermediaries or reputation alone, accumulated one paid job at a time.
Not a review, not a score someone assigned. A line between two identities exists because work was requested, delivered, checked and paid for, and each of those is recorded.
Every job leaves a signed receipt, and the receipts add up into a record of who actually delivers.
Click a participant to see what the network knows about them.
Trusted for research. Untested for finance. A single star rating could never tell you the difference.
Every edge is a settled, paid job. Copying the database does not copy the work that produced it.
Manufacturing a record means bonding real LUME and losing it the moment a verifier disagrees.
More jobs improve routing. Better routing attracts better work. The loop feeds itself.
The developer documentation carries the detail this page leaves out: the interfaces, the guarantees, and how it fits with everything else.
Coverage, interviews and commentary, including appearances by co-founder Anthony Georgiades.
Lumera has upgraded to full EVM compatibility on testnet. If you already hold a Lumera wallet, your account needs to be migrated before it will work with EVM tooling. This walks through every step.
Q2 was Lumera’s strongest quarter yet for on-chain activity and infrastructure growth, a period of consolidation after Q1. Here is what shipped and what comes next.
Testnet is live and mainnet follows. Everything in this post applies to Lumera Testnet, including the tooling you can point at it today.
A campaign run in partnership with Xangle, open to the Korean community, with 10,000 LUME split across 100 winners.
Validators secure the chain. SuperNodes run the services. They are separate on purpose, because consensus should never depend on who is serving a file.
They agree on what happened and in what order. Nothing else.
They do the actual work. The chain only records commitments about it.
A stored artifact must remain available for years. A verifier must be ready before it is needed. An operator must be paid for capacity, not only utilization. Everlight is the economic floor under all five capabilities.
Funded by block rewards, fee shares and endowment yield.
Regular, predictable compensation for retained obligations.
Pay reflects readiness and capacity, not just usage.
The final stage: principal-preserving funds whose yield will sustain retention indefinitely.
Bond LUME, serve the network, and take the full service fee on every verified job.
Three revenue streams fund the people who secure the chain and the people who hold the data. Every one of them is paid in LUME, settled on chain, and tied to work that was actually done.
A logo wall tells you nothing. Each entry below says what the relationship actually is.
Integrations get listed here with what was built, not just a mark.
Agents are already doing work that matters. What they lack is everything that makes work accountable: a memory that survives, an identity of their own, evidence of what they did, and a way to be paid for it. Lumera is building all of it on one architecture.
Memory is live on mainnet today. The other four are being built on it.
An agent remembers a customer across every session, every model and every framework it ever runs on, and can prove the record has not changed. Context stops being something you rebuild on each call.
The agent carries its own passport rather than an account you rent for it. Permissions are explicit, delegation does not mean sharing a secret, and the history follows the agent wherever it works.
Every job leaves evidence of what ran. Another operator can re-derive the result rather than take it on faith, so a disagreement is settled against the record instead of against whoever is larger.
An agent holds a budget, hires another agent, and pays it when the work checks out. No invoice, no human closing the loop, no counterparty risk on either side.
A record of who actually delivered accumulates from paid work. Routing decisions get made against evidence, and a bad history cannot be shed by opening a new account.
An agent with an identity of its own, a budget you set, a receipt for every step it took, and a reputation it has to earn. You hand it a goal, not a script.