Make AI
ask first.
An open protocol for consent between humans and AI systems.
Today AI trains on your data, learns from your words, and acts on your behalf without a way to ask, prove permission, or pay you back. LLMConsent is the missing layer: verifiable consent, attribution, and compensation, the way HTTP is a layer, owned by no one.
AI runs on people who never agreed.
Every platform reinvents consent in private, if at all. The result is the same everywhere.
A conforming implementation:
Five properties, written the way a standard writes them. Non‑negotiable.
Be decentralized
No single entity controls the protocol. No admin keys, no central authority that can freeze, seize, or override a person's consent.
Keep people sovereign
People own and control their data and their AI representations: what gets used, by which models, for how long, and at what price.
Interoperate
A common language any AI system can implement. Like HTTP or TCP/IP, it works across platforms, models, and vendors.
Be verifiable
Consent is cryptographically signed and checkable. Permission and attribution can be proven, not merely claimed.
Settle fairly
Compensation is built in. When data or a digital twin is used, value can flow back to the person it belongs to.
Five steps. Same for everyone.
Every interaction runs the same verifiable loop, whoever is asking.
Four standards. One stack.
Each one builds on the last, from a single consent token up to shared memory across agents.
Core Consent
Cryptographically-verifiable consent tokens that scope how data may be used for training, inference, agent actions, and memory, with built-in attribution limits, expiry, economics, and revocation.
foundation, no dependencies · v0.1.0Digital Twin
User-owned, persistent AI representations that evolve across interactions, with fine-grained access levels and privacy controls, so personalization never means losing ownership.
requires LCS-001 · v0.1.0Agent Permissions
Capability-based security for autonomous agents: what they may do, how much they may spend, who they may delegate to, and how it is all revoked and audited.
requires LCS-001, LCS-002 · v0.1.0Cross-Agent Memory
Consent-governed memory pools that let agents share context across systems, so experiences carry over without surrendering control of what's remembered.
requires LCS-001, LCS-002, LCS-003 · v0.1.0One check before you use the data.
Implement the consent layer in the language you already build in.
// Check user consent before training
function trainModel(bytes32 dataHash) external {
require(
llmConsent.checkConsent(msg.sender, dataHash, PERMISSION_TRAIN),
"Training consent required"
);
// ...proceed with training
}# Verify consent before processing a request
def process_request(user_data):
if not llm_consent.verify(user_data.hash, Permission.INFERENCE):
raise ConsentRequired("User consent needed")
return ai_model.process(user_data)// A user grants scoped, inference-only consent
const consent = await llmConsent.grantConsent({
dataHash: myDataHash,
permissions: PERMISSION_INFER,
duration: 86400, // 24 hours
models: ["any"], // or specific model ids
});Reference SDKs are unwritten. claim one
A protocol, not a platform.
LLMConsent is community-owned and developed in the open, modeled on the IETF, W3C, and the BIP/EIP improvement-proposal processes. Decisions are made by rough consensus, not by a company.
No single owner
The protocol is community-owned. No admin keys, no ability to freeze or seize.
Transparent process
All decisions happen in public, with open participation for anyone.
Rough consensus
Standards advance by general agreement and working code, not by vote.
Protocol, not platform
We define standards and encourage many independent implementations.
A complete proposal, championed by its author and open for feedback.
Formal community review of at least 30 days, with a maintainer sponsor.
A final 14-day window for objections once rough consensus is reached.
An accepted standard with a reference implementation. Immutable after.
Where it is heading.
v1.0 is the foundation. The path ahead deepens interoperability, scale, and next‑generation cryptography.
Foundation
- Core consent tokens & verification
- User data sovereignty
- Built-in compensation model
- The four core standards (LCS-001 → 004)
Interoperability
- Cross-chain support
- Additional DID methods
- Enhanced privacy (ZK proofs)
- Batch operations
Scalability
- Layer-2 scaling
- State channels
- Compressed tokens
Next generation
- Post-quantum cryptography
- AI-native consensus
- Autonomous negotiation
Build the standard with us.
This is the ground floor. v1.0 just landed, the four standards are still Draft, and the reference SDKs are unwritten. Whoever shows up now gets to shape what consent for AI actually means. Your next pull request could define LCS‑005.
Not ready to contribute? Stay in the loop.
One short email when a standard ships or a new proposal opens. No spam, unsubscribe anytime.
Implement an SDK
Build the Python, JS, or Solidity reference client. The interfaces are specified, the code is not.
github.com/llmconsent → shape the specPropose LCS-005
See a gap in the protocol? Open an LCP through the public proposal process and make the case.
read the process → reviewPressure-test the drafts
Find the holes. Review open proposals, file issues, and help reach rough consensus.
open issues → communityShow up early
Introduce yourself, say what you want to build, and help set the direction from day one.
join the discord →