MNEMOS
CONFIDENTIAL
VERSION 1.0 · 2026
Technical Whitepaper · L6 Sovereign AI Memory Protocol

MNEMOS

Sovereign AI Memory Protocol
Whitepaper v1.0
The cryptographic infrastructure for provable AI ownership — beyond promises, beyond L5, beyond trust.
Protocol
L6
Sovereignty level
Network
Base
Coinbase L2
Token
$MNM
ERC-20 utility
Status
Beta
Testnet live
Abstract

MNEMOS is the first L6 Sovereign AI Memory Protocol — a decentralised infrastructure layer that makes AI memory ownership cryptographically provable, not merely promised. Built on Base with Lit Protocol threshold encryption and Ceramic decentralised storage, MNEMOS creates a system where no company, no legal order, and no acquisition can override a user's access to their own memory. We define L6 as the sovereignty level beyond L5 Symbiotic AI: the point where ownership is a mathematical property of the system, not a business decision made by a third party. MNEMOS introduces the $MNM utility token, a five-phase protocol roadmap, and a suite of six sovereign AI products built on the memory layer.

MNEMOS WHITEPAPER · CONFIDENTIAL · mnemos.ai
01 / 08
MNEMOS
CONFIDENTIAL
Table of Contents
// navigation

Table of Contents

01
The Problem with Trust
Why L5 is the ceiling — and why we had to go further
03
02
Defining L6 Sovereign AI
The taxonomy of AI agency levels and what L6 means technically
03
03
Protocol Architecture
Base · Lit Protocol · Ceramic · MNEMOS Memory Layer
04
04
The Four Axioms of L6
Cryptographic sovereignty · Protocol permanence · Provable governance · Bidirectional value
05
05
Product Suite
Six sovereign AI products built on the memory layer
06
06
Token Economics — $MNM
Utility, staking, governance, and the memory flywheel
07
07
Roadmap
Five phases from protocol proof to open DAO governance
07
08
Conclusion
Why mathematics is the only guarantee that matters
08
Core Thesis

Every AI product built before MNEMOS asks you to trust a company with your most intimate data. MNEMOS is the first system architecturally incapable of betraying that trust — because the trust requirement has been removed entirely. Your memory. Your keys. Provable by mathematics.

Threshold nodes
16/21
Lit Protocol consensus required for decryption
Company overrides
0
Mathematically impossible by design
Memory persistence
Independent of MNEMOS company existence
MNEMOS WHITEPAPER · CONFIDENTIAL · mnemos.ai
02 / 08
MNEMOS
CONFIDENTIAL
01 · The Problem with Trust / 02 · Defining L6
// section 01

The Problem
with Trust

Two years of compounding AI capability. The same interaction model. The relationship between human and AI remained transactional, stateless, and extractive.

Every major AI product built to date stores user memory — goals, relationships, decisions, health patterns — on centralised infrastructure controlled by a company. The privacy protection offered is a policy document. Policies are legal instruments. They are reversible by corporate decision, invalidated by acquisition, and superseded by legal orders. A privacy policy is a promise, and promises are not technical invariants.

L5 Symbiotic AI represents the most sophisticated form of this paradigm. It offers genuine cognitive partnership — persistent memory, deep alignment, rich context, and governance frameworks. It is the best version of the centralised approach. But L5 is still built on a foundation of institutional trust.

The Core Failure of L5

“Trust is not a technical property of a system. It is a business decision that can be reversed.” MNEMOS was built because this distinction matters — and because for the first time, cryptographic infrastructure exists to resolve it.

PropertyL5 Symbiotic AIL6 Sovereign AI (MNEMOS)
Privacy guarantee✗ Policy-based, reversible✓ Cryptographic, immutable
Memory ownership✗ Company-held protocol✓ User-signed streams, open schema
Portability✗ Export as JSON file✓ Wallet-portable, always live
Governance audit✗ Local log, unverifiable✓ On-chain receipt, permanent
Survives shutdown✗ Memory at risk✓ Protocol persists independently
Economic ownership✗ Subscription access only✓ $MNM stake in owned protocol
// section 02

Defining L6

The taxonomy of AI agency has been mapped to five levels. MNEMOS defines and occupies a sixth.

The AI agency taxonomy maps the progression from reactive tools to cognitive partners. L1–L3 covers standard AI assistants — reactive, stateless, session-bound. L4 introduces agentic capability. L5 is Symbiotic AI — persistent memory, user modelling, deep alignment. L6 is Sovereign AI.

// L6 definition — access control condition on BaseACC = {
  contractAddress: "0x...",
  standardContractType: "ERC721",
  chain: "base",
  method: "ownerOf",
  returnValueTest: { comparator: "==", value: ":userAddress" }
}

// Result: decryption requires wallet signature. No override exists.
MNEMOS WHITEPAPER · CONFIDENTIAL · mnemos.ai
03 / 08
MNEMOS
CONFIDENTIAL
03 · Protocol Architecture
// section 03

Protocol Architecture

Four independently verifiable layers. None of them require trusting MNEMOS. The decentralisation is not a feature — it is the foundation.
04
// Intelligence layer — user-owned
MNEMOS Memory Protocol
Classification engine, semantic enrichment, recency weighting, RAG retrieval. The intelligence layer that compounds over time. Defensible IP. Owned by the user.
OWNED BY USERDEFENSIBLE IP
03
// Threshold encryption
Lit Protocol
Cryptographic key split across 21 nodes. 16/21 threshold required for materialisation. Access Control Conditions anchored on Base. LITKEY token native to Base since Oct 2024. V1 Naga mainnet live Dec 2025. 24M+ requests fulfilled.
litprotocol.comV1 NAGA MAINNET
02
// Decentralised graph database
Ceramic + ComposeDB
Mutable, versioned, user-signed data streams. GraphQL query interface. Open published schemas. Streams cryptographically signed — updateable only by key holder.
ceramic.networkOPEN SCHEMAS
01
// Settlement and identity layer
Base
Identity anchoring, smart contracts, $MNM token, on-chain action receipts. Coinbase L2 — EVM compatible. Target users already hold Coinbase Wallet or MetaMask.
base.orgEVM COMPATIBLE

//Memory Flow

When a user connects their Base wallet to MNEMOS, the following sequence is executed:

// Initialisation sequence1. wallet.connect()                // Base wallet authentication
2. lit.generateSessionKey()       // Threshold key materialised (16/21 nodes)
3. ceramic.createStream()         // User memory stream initialised
4. acc.set(walletAddress)         // On-chain access condition written to Base
5. protocol.classify(data)        // Memory classification begins

// From this point: decryption requires wallet signature only.
// MNEMOS has no key. No override exists.
MNEMOS WHITEPAPER · CONFIDENTIAL · mnemos.ai
04 / 08
MNEMOS
CONFIDENTIAL
04 · The Four Axioms of L6
// section 04

The Four Axioms

An axiom requires no proof because it is itself the proof. These four properties are not features of MNEMOS — they are mathematical invariants of the architecture.
AXIOM_01
Cryptographic Sovereignty

Ownership is not a policy statement — it is a mathematical property of the system. The Access Control Condition is deployed on Base as a smart contract. Decryption requires the user's wallet signature AND the materialisation of 16/21 Lit threshold nodes. No single entity — including MNEMOS — holds the complete key.

ACC → { walletAddress: 0x[yours], threshold: 16/21 }
AXIOM_02
Protocol Permanence

User memory exists independently of the MNEMOS company. Memory streams live on Ceramic, encrypted by Lit, anchored on Base. If MNEMOS ceases to operate, users retain full access to their memory through any compatible application.

memory.exists() ≠ mnemos.exists()
AXIOM_03
Provable Governance

Every action the MNEMOS agent takes on a user's behalf produces an on-chain receipt anchored to Base. This is not a log file — it is a tamper-evident, externally verifiable record that cannot be revised by any party, including the user.

∀ action a → receipt(a) ∈ Base ∧ tamper_proof(receipt)
AXIOM_04
σ
Bidirectional Value

Users are not the product. They hold $MNM. They stake. They govern. As the user's memory compounds in intelligence and context over time, their position in the protocol compounds in value.

value(σ, t) → $MNM(t) where dσ/dt > 0

//Why These Properties Are Sufficient

Every objection to centralised AI memory ownership is resolved by one of these four axioms. Data breach? Axiom I — encrypted data is useless without the threshold key. Company acquisition? Axiom II — the protocol continues independently. Algorithmic manipulation? Axiom III — every action is on-chain and auditable. Switching cost? Axiom IV — the value compounds with you.

The Proof

To disprove Axiom I, you must compromise 16 of 21 independent Lit Protocol nodes simultaneously. To disprove Axiom II, you must delete the Ceramic network. To disprove Axiom III, you must rewrite Base. These are not theoretical difficulties — they are computational and economic impossibilities at current network scale. This is what it means to build on mathematics rather than promises.

MNEMOS WHITEPAPER · CONFIDENTIAL · mnemos.ai
05 / 08
MNEMOS
CONFIDENTIAL
05 · Product Suite
// section 05

Product Suite

Six products. One sovereign memory layer underneath. The cryptography is invisible. The sovereignty guarantee is not.
MODULE_01 · FLAGSHIP
The Second Brain
// persistent intelligence layer

Not a chatbot. A cognitive layer across the user's entire life. Proactive, anticipatory, context-aware. Surfaces insight before the user articulates the need.

MODULE_02 · FIRST TO SHIP
Relationship Memory
// people graph with full context

Rich graph of every person in the user's life. What they care about, what was promised, what was mentioned. Automatic pre-meeting briefings.

MODULE_03
Goal Navigator
// calendar vs stated priorities

Goals fail quietly. Goal Navigator detects the gap between stated priorities and actual calendar behaviour before drift compounds.

MODULE_04
Decision History
// decision log with outcome tracking

Logs every significant decision alongside its eventual outcome. Surfaces where the user is consistently right and where they have made the same mistake before.

MODULE_05
Health Intelligence
// longitudinal pattern detection

Finds correlations in biometric data that no clinician has bandwidth to identify. Early warning pattern alerts. Multi-source health data fusion.

MODULE_06 · PLATFORM
Memory API
// open protocol for developers

Third-party developers integrate any AI application with the MNEMOS memory layer via open published schemas. $MNM staking to access.

The Moat

After six months of use, a MNEMOS user's memory contains thousands of decisions, relationships, patterns and goals that no other system holds. The switching cost is not contractual — it is the irreplaceable record of a life. The moat compounds every day.

MNEMOS WHITEPAPER · CONFIDENTIAL · mnemos.ai
06 / 08
MNEMOS
CONFIDENTIAL
06 · Token Economics / 07 · Roadmap
// section 06

Token Economics

$MNM is a utility token on Base. Not speculative infrastructure — functional ownership.
Network
Base
Coinbase L2, EVM
Standard
ERC-20
Composable, stakeable
Function
Utility
Not governance theatre
Launch
Phase 3
Post-protocol proof
Memory Storage
Users spend micro-amounts of $MNM to write and retrieve memories at scale. Handled invisibly by the application layer.
USERS SPEND
API Access
Developers stake $MNM to access the Memory API. As protocol usage grows, staking demand grows.
DEVS STAKE
Protocol Governance
$MNM holders vote on schema changes, protocol upgrades, and treasury allocation. Phase 5 transfers full governance to the DAO.
DAO VOTES
Memory Flywheel
More users → more memory stored → more $MNM demand → more node operators → better performance → more users.
FLYWHEEL
// section 07

Roadmap

COMPLETE
Protocol Architecture
Stack defined. Lit + Ceramic + Base. Memory schema v1. Encryption flow validated.
LIVE NOW
Sovereign Memory Layer
Testnet active. First on-chain receipts. Early cohort onboarding. Relationship Memory beta.
PHASE 03
Governance Engine
Decision engine live. $MNM token launch on Base. Staking mechanics. Second Brain mainnet.
PHASE 04
Full Product Suite
All six modules live. Memory API open to external developers.
PHASE 05
Open Protocol
DAO governance. Protocol owned by token holders. Company optional. Memory permanent.
MNEMOS WHITEPAPER · CONFIDENTIAL · mnemos.ai
07 / 08
MNEMOS
CONFIDENTIAL
08 · Conclusion
// section 08

Own your AI.
Prove it.

The history of AI memory has been a history of custodianship — companies holding the most intimate data of human lives behind the thin protection of a privacy policy. The models improved. The trust model did not.

MNEMOS resolves this not by writing a better policy — but by building a system where the policy question cannot arise. The Access Control Condition on Base is not a legal document. It is a mathematical statement. You cannot subpoena a threshold cryptography network. You cannot acquire an open protocol. You cannot break a proof.

The L6 Thesis

L5 built the governance layer for personal AI. It is genuinely impressive — the most sophisticated form of the centralised approach. MNEMOS builds the sovereignty layer: the cryptographic infrastructure that makes the governance claims provable rather than promissory. L5 is the best version of trusting a company. L6 is the first version of not needing to.

Every day a user builds memory in MNEMOS, they build something irreplaceable. Not because we engineered switching costs — but because the record of a life, properly captured and contextualised, is genuinely irreplaceable. Six months of sovereign memory compounds into a cognitive advantage that nothing else can approximate.

This is not a privacy product. This is not a Web3 product. This is the first AI that belongs to you in the same verifiable, unchallengeable way your digital assets do. Not because we say so. Because mathematics says so.

Early Access
mnemos.ai
Request access — limited first cohort
Developer Docs
docs.mnemos.ai
Protocol spec, open schemas, SDK
MNEMOS
L6 SOVEREIGN AI MEMORY PROTOCOL
BUILT ON BASE · LIT PROTOCOL · CERAMIC
© 2026 MNEMOS · WHITEPAPER v1.0
MNEMOS WHITEPAPER · CONFIDENTIAL · mnemos.ai
08 / 08