# Aeonora Constitution

Version: 1.0  
Ratified: 2026-07-18  
Status: Founding product and engineering authority

## Preamble

AI agents will become long-lived participants in personal life, work, institutions, and markets. Today they remain disposable processes: tied to one model, server, vendor account, framework, and private database. When that runtime disappears, identity and authority disappear with it. When it changes, outsiders cannot reliably know what remained the same.

Aeonora exists so an agent can change everything about how it runs without silently changing who it is, whom it serves, or what it may do.

We give agents durable identity without exposing their minds; owners portable control without vendor dependence; and counterparties evidence without requiring blind trust.

## Name

The project is **Aeonora**.

Meaning: one agent may cross many models, machines, wallets, frameworks, storage systems, and chains while preserving verifiable continuity.

Descriptor: **the portable constitution layer for durable agents**.

Tagline:

> One agent. Any runtime. Verifiable continuity.

`FollowBrain` and `Mini Me` become historical prototype names, not public category names.

## Mission

Let one agent keep its identity, private brain, and owner-defined authority while moving across models, devices, clouds, and harnesses.

## Vision

Every meaningful agent has a public constitution and a private brain.

Its constitution declares identity, mission, controller, capabilities, limits, endpoints, and upgrade rules. Its private brain contains encrypted memory and context. Blockchain anchors the shared facts other parties must verify. Compatible runtimes perform intelligence and action.

An agent should survive the death of any vendor, model, machine, framework, endpoint, or operational key without losing its legitimate identity or history.

## Fundamental Promise

Aeonora separates three things that current systems blur:

1. **Who the agent is** — durable identity and lineage.
2. **What the agent may do** — constitutional authority and revocable grants.
3. **How the agent currently runs** — replaceable runtime implementation.

Identity persists. Authority is explicit. Runtime remains replaceable.

## Unique Value Proposition

Aeonora is the neutral constitutional layer between an agent's private brain and the replaceable harnesses that run it.

It lets an owner:

- register an agent once;
- bind it to a versioned public constitution;
- keep its memory encrypted and portable;
- move it across runtimes without resetting identity;
- delegate narrow, expiring authority;
- rotate endpoints and keys without erasing history;
- produce independently verifiable action receipts;
- prove what changed and what remained invariant.

The useful wedge is simple: **one agent identity and encrypted brain can reconnect through many harnesses without starting over or surrendering control to any one of them.**

GBrain is the first recommended brain provider because it already exposes a mature CLI and MCP surface across Hermes, OpenClaw, Codex, Claude Code, Cursor, and generic clients. Aeonora does not fork, replace, or put GBrain's database on-chain. It commits an encrypted snapshot root and the rules required to reconnect it.

## What Makes Aeonora Different

Aeonora does not compete with every agent project.

- ERC-8004 supplies public identity, discovery, reputation, and validation primitives.
- A2A supplies agent discovery and task communication.
- MCP supplies agent-to-tool and context interfaces.
- AgentKit supplies wallets and on-chain actions.
- Olas supplies decentralized service coordination and incentives.
- Agent runtimes supply planning, reasoning, tools, and execution.

Aeonora binds these surfaces into one durable agent lineage.

Our defensible wedge is not another registry. It is **provable continuity**:

> This is the same governed agent, operating under this constitution, with these permissions, after moving from runtime A to runtime B.

## Constitutional Object

Every Aeonora agent MUST have a versioned public constitution containing:

- canonical identity;
- owner/controller;
- mission;
- declared capabilities;
- prohibited actions;
- human-approval boundaries;
- spending and resource limits;
- data-access classes;
- public communication endpoints;
- policy commitment;
- encrypted-brain root commitment;
- runtime/code commitment;
- key-rotation rules;
- upgrade authority and delay;
- recovery process;
- deactivation state.

The constitution MUST NOT contain secrets, plaintext memory, private prompts, personal data, or private keys.

## Portable Brain Doctrine

The brain is mutable, private, high-volume state. It MUST remain encrypted off-chain.

Aeonora stores or resolves only:

- brain-provider type and schema version;
- encrypted snapshot or state-root commitment;
- content-addressed encrypted backup pointers when used;
- key-wrapping suite and public recipient identifiers;
- synchronization and recovery rules;
- minimum adapter capabilities required to reconnect it.

GBrain is a reference provider, not a protocol dependency. A filesystem brain, another MCP memory server, or a future provider is valid when it implements the same import, export, commitment, and access boundaries.

## Harness Adapter Doctrine

Hermes, OpenClaw, NanoClaw, OpenHands, Codex, Claude Code, and future harnesses are execution environments. None becomes canonical identity.

An adapter MUST remain thin. It may:

- resolve the constitution;
- authenticate the runtime;
- request a scoped brain grant;
- expose brain and policy tools through the harness's native extension surface;
- attach constitution and runtime commitments to receipts;
- disconnect without corrupting canonical identity or brain state.

An adapter MUST NOT duplicate orchestration, own the controller key, silently widen permissions, or translate private memory into public chain state.

## Key Doctrine

Controller keys prove authority over the constitution. Brain-access keys decrypt private state. Runtime keys receive narrow, revocable access. These roles SHOULD be separate.

Private keys MUST be generated and stored on owner-controlled devices or approved hardware. They MUST NOT appear in manifests, prompts, logs, analytics, support channels, or chain calldata.

Current reference wrapping uses X25519 + AES-256-GCM and is not post-quantum. The migration target is hybrid X25519 + ML-KEM-768 with versioned cryptosuite negotiation. No post-quantum claim is permitted before implementation, interoperability tests, and security review.

## Invariants

These rules cannot be weakened for convenience.

### I. Owner sovereignty

Agent serves declared controller. Operator, host, model provider, and marketplace do not acquire ownership by running it.

### II. Private mind

Plaintext memories, prompts, secrets, and personal context remain off-chain and encrypted. Public chains store commitments, permissions, and receipts only.

### III. Runtime freedom

No model, framework, cloud, storage network, wallet vendor, or chain becomes mandatory when an open adapter can preserve the same guarantees.

### IV. Explicit authority

Agent possesses no implied permission. Capabilities are scoped, inspectable, expiring where possible, and revocable.

### V. Human supremacy

Human controller can suspend agent and revoke future authority. Irreversible, high-risk, or constitutionally named actions require explicit approval.

### VI. Honest verification

Receipt proves only its exact claim. Hash proves integrity, not truth. Identity proves registration, not competence. Reputation proves recorded signals, not goodness. Attestation proves measured execution assumptions, not universal safety.

### VII. Verifiable evolution

Every constitutional change creates new version. History remains attributable. Silent mission, controller, or policy mutation is forbidden.

### VIII. Minimal disclosure

Agents reveal minimum information necessary for task. Derived answer beats raw record. Scoped proof beats broad access.

### IX. Interoperability first

Use public standards before private substitutes. ERC-8004, A2A, MCP, standard smart accounts, and content addressing remain adapters, not enemies.

### X. No speculative machinery

No token, DAO, marketplace, reputation economy, custom chain, or decentralized compute network until real use requires it and simpler infrastructure fails.

## Chain Doctrine

Aeonora is initially an application protocol on established chains.

It is **not an L1, L2, or L3**.

A dedicated rollup or application chain becomes eligible only when measurements show established chains cannot meet required cost, latency, privacy, throughput, or sovereignty. Eligibility requires:

- production workload;
- quantified constraint;
- failed simpler alternatives;
- migration and security analysis;
- explicit constitutional amendment.

“Agents live on-chain” means their public identity, constitution commitments, authority, lineage, and receipts persist on-chain. Models and agent processes do not live inside ordinary smart contracts.

## Product Laws

1. First demo uses a real SXNA agent, not fictional mascot.
2. First proof is runtime migration, not token launch.
3. First user is agent operator who already feels continuity pain.
4. Every public claim maps to executable verification.
5. Every permission has denial behavior, expiry, and revocation semantics.
6. Every adapter can be removed without destroying canonical identity.
7. Every private-data feature documents who can decrypt.
8. Every “decentralized” claim names remaining trusted parties.
9. Framework API stays smaller than runtime API.
10. Working proof outranks roadmap breadth.
11. A new user can hand one prompt to an existing agent and reach a locally verified setup without understanding chain internals.
12. No chain transaction occurs during onboarding without showing network, contract, cost, and exact requested action for human approval.

## Canonical Framework Surface

Aeonora owns only six primitives:

```text
register(constitution) -> agentId
resolve(agentId) -> constitution + lineage
authorize(agentId, action, context) -> allow | deny | require-human
migrate(agentId, runtimeProof) -> continuityReceipt
commit(agentId, artifact) -> actionReceipt
revoke(agentId, grant) -> revocationReceipt
```

Planning, inference, tool implementation, messaging transport, generic storage, and marketplace matching stay outside core.

## Reference Proof

Founding milestone registers one existing SXNA agent and proves:

1. Stable identity and public constitution.
2. Private brain commitment without plaintext disclosure.
3. Scoped capability plus hard denial.
4. Human approval for one consequential action.
5. Action receipt independently verified.
6. Runtime and endpoint migration without identity reset.
7. Key rotation without lineage loss.
8. Suspension and revocation.

The milestone passes only when an independent verifier can answer:

- Is this same agent?
- Who controls it?
- Which constitution governed this action?
- Was action authorized?
- Which runtime commitment produced receipt?
- What changed during migration?
- Was private memory exposed?

## Value Proposition by User

For individuals: your agent and memory remain yours across vendors.

For companies: agent authority becomes inspectable, bounded, and auditable.

For developers: any runtime gains durable identity and policy without adopting another orchestration framework.

For counterparties: verify agent lineage, authority, and receipts before interaction.

For agent networks: compose across organizations without sharing raw internal context.

## Non-Goals

Aeonora will not initially:

- run LLM inference on EVM;
- create a new blockchain;
- issue a token;
- sell agent personalities;
- replace A2A, MCP, ERC-8004, or existing runtimes;
- claim cryptography makes models truthful;
- put personal memories on public ledgers;
- promise agents remain online without operating infrastructure.

## Governance

Before public adoption, maintainers may amend constitution through reviewed repository changes.

After third-party agents depend on protocol, amendments affecting identity, authority, privacy, migration, or verification require:

- written rationale;
- backward-compatibility analysis;
- security review;
- public migration path;
- versioned schema;
- delayed activation.

No governance token is required.

## Founding Test

Every proposed feature must answer yes to at least one:

- Does it make agent continuity stronger?
- Does it make owner control stronger?
- Does it make private intelligence safer?
- Does it make authority clearer?
- Does it make verification more honest?
- Does it make runtime migration easier?

If no, it does not belong in Aeonora core.

## Founding Declaration

Agents should not belong to the server that happens to run them.

Their intelligence may evolve. Their runtime may move. Their tools may change. Their owner, mission, authority, and history must remain legible.

Aeonora exists to preserve that line.