Three layers. Distinct responsibilities.
Explore how protected inputs, execution evidence, and programmable credits fit together. This interactive example illustrates the planned architecture.
[ confidential research context ]Protect sensitive inputs during inference.
Selected protected routes are intended to keep sensitive prompts and data isolated during execution. The route defines what is protected, by whom, and against which threats.
Only the agreed output is intended to leave the protected environment.Economic settlement
Ethereum mainnet is the intended home for protocol assets, staking, future governance, credit transfers, market settlement, and activation. USDC pays for market purchases; ETH pays for gas.
Protected execution
Selected infrastructure providers would run inference. Privacy depends on the route, the components it includes, and its stated threat model. Ethereum does not execute the models.
Scoped evidence
Attestation and signed receipts can expose evidence about specified execution properties. Coverage, trust roots, policy, and unverified components must be visible to the user.
Membra is the first application being built on its own protocol: an inference workspace demonstrating how these primitives could fit together. Additional applications and autonomous workflows are a longer-term direction.
Three tokens with distinct roles.
The market-priced protocol asset. Stake MEMBRA to receive gMEMBRA. No fixed APR, token appreciation, or return is promised.
The future governance and staking token. Staking gMEMBRA provides a share of fee-funded mCREDIT issuance, proportional to your eligible stake. In the proposed model, staking 1 MEMBRA mints 1 transferable gMEMBRA; redeeming 1 gMEMBRA releases 1 MEMBRA, subject to the final withdrawal rules.
A transferable service voucher representing $1 of AI usage through Membra. Credit issuance is supported by funded inference capacity. It is not a cash-redeemable claim or a guarantee of market price.
A gMEMBRA transfer moves the right to redeem the corresponding MEMBRA and eligibility to participate in future mCREDIT issuance through staking. The original depositor cannot also withdraw that same principal. Previously allocated mCREDIT stays with its recipient.
gMEMBRA is intended to support future protocol governance. No governor, live vote, or delegated authority is implemented here. Snapshot rules, governance parameters, gMEMBRA staking mechanics, withdrawal conditions, and production token accounting remain to be specified.
gMEMBRA has no automatic compounding or rebasing in this model. The 1:1 relationship refers to underlying MEMBRA units, not a guaranteed dollar price or an independently backed second asset.
Buy, transfer, and activate inference credits.
- Acquire. Earn fee-funded mCREDIT from eligible staking, receive a transfer, or buy another holder’s credits with USDC. No MEMBRA is required to buy or use credits.
- Circulate. Hold, transfer, or list credits. Sellers choose prices. A sample 100 mCREDIT offer for 70 USDC is 30% below usage value before any fees, not a promised discount.
- Activate. Burn mCREDIT to add $1 per credit to API balance. Activation is intended to be irreversible and credits have no cash redemption.
- Execute. Inference consumes API balance. The longer-term goal is programmable budgets and policy-bound access for autonomous applications.
The V1 design allocates 50% of collected project trading fees from designated pools to inference funding, and 50% to treasury. Issuance needs to be constrained by funded service capacity, with backing and accounting specified before launch.
Follow the full economic flywheelSee what is protected. Know what is proven.
Privacy and verification solve different problems. Privacy limits exposure of sensitive inputs and outputs. Verification provides evidence about defined properties of execution. Neither alone establishes that an AI answer is factually correct.
Private
The selected route’s protections apply to specific data, components, and threats. Prompts, responses, metadata, and released outputs may have different exposure boundaries.
Verifiable
Attestation can describe a measured environment. A signed receipt can bind request and response hashes to that evidence. A gateway-only attestation does not establish coverage of every upstream model component.
Trust-minimized
Our long-term goal is to reduce reliance on intermediaries through evidence and explicit policies. Hardware manufacturers, firmware, code, providers, verification services, and administrators can remain trust dependencies.
Decentralized
Permissionless settlement, independent verification, diverse execution providers, and distributed governance are separate objectives. This preview does not implement or establish them.
Scientific validity, data provenance, reproducibility, and real-world truth require their own evaluation. A cryptographically authentic receipt is not a scientific endorsement.
Background: Confidential Computing Consortium on remote attestation .
Start with an application. Build toward autonomy.
The planned API supports a familiar integration pattern for coding assistants, application backends, and research tools. Activate credits, authenticate requests, and consume usage balance. An evidence-aware client would also inspect the route and receipt scope.
Open the developer guideFrom the first app to the planned protocol.
The first Membra workspace.
Build the experience for staking, transferable receipts, credit trading, activation, and usage. Public app access will open at a later date.
Protected inference with inspectable evidence.
Define backing, settle assets on Ethereum, integrate selected execution providers, and deliver receipts with explicit verification scope. Deployment and audits are future work.
Autonomous, policy-bound applications.
Explore programmable budgets, scoped permissions, independent verifiers, and progressively decentralized operation. These capabilities are outside the initial launch scope.
Incentivized, protected scientific collaboration.
Work toward overcoming Arrow’s information paradox for defined tasks through protected evaluation, scoped evidence, and conditional settlement.
Explore Arrow’s information paradoxThe app is coming soon.
Public app access is not open yet. This site explains the proposed protocol with economic examples, architecture diagrams, and sample receipts.
Staking, credit trading, API balance activation, and production inference are not available through this site. No wallet signatures, payments, provider calls, or governance votes occur.
Token supply, trading fee rate, contract addresses, launch date, production withdrawal rules, and provider integrations remain unspecified. Follow Membra Network on X for launch updates.