Skip to content

manifesto.md

Latest commit

 

History

History
99 lines (60 loc) · 2.77 KB

manifesto.md

File metadata and controls

99 lines (60 loc) · 2.77 KB
Raw

🌱 The Hussh Manifesto

Consent is not a checkbox.
It’s a contract, a signal, and a programmable boundary.

Visual Context

Canonical visual owner: consent-protocol. Use that map for the top-down system view; this page is the narrower detail beneath it.

🧠 Why We Exist

The Hussh Consent MCP surface was built to give humans control over their data in an AI-powered world.

Most AI agents today are trained to maximize engagement, extract value, or serve the platform.
We believe agents should serve the person — and only when asked to.

That’s why we built a programmable trust layer that:

  • Requires explicit permission from the user
  • Can be verified cryptographically
  • Can be revoked instantly
  • Can’t be faked

🔐 Design Principles

1. Consent is First-Class

Agents do nothing until permission is granted.
We use signed tokens and scoped access — not vague “I agree” buttons.

2. Trust Must Be Programmable

We enable delegation of access through verifiable TrustLinks, not hardcoded relationships.
One agent can trust another, but only within defined scopes and durations.

3. Data is Vaulted

All sensitive information is encrypted using AES-256-GCM.
Only agents with valid keys and tokens can see the data — and even then, only within scope.

4. Everything is Auditable

Tokens and links are signed, inspectable, and time-bound.
We assume your system will be inspected — by humans or code.

5. Modularity Wins

Inspired by operons in biology, we design agent actions to be small, reusable, and testable.
Agents should be swappable. Behaviors should be composable.

🔍 What We Don’t Do

  • ❌ We don’t trust the platform by default
  • ❌ We don’t store plaintext user data
  • ❌ We don’t assume “user signed in” = user gave consent
  • ❌ We don’t allow agents to bypass protocol logic

⚖️ Real Consent is Verifiable

If your agent can’t show:

  • ✅ A signed consent token (HushhConsentToken in the compatibility type surface)
  • ✅ A matched ConsentScope
  • ✅ A validated TrustLink (if delegated)

…then it doesn’t have permission. And it shouldn’t act.

💡 What We Hope You’ll Build

  • 🧠 Agents that summarize, organize, and protect your life
  • 🤖 Teams of agents that collaborate with scoped trust
  • 🧬 AI that’s modular, permissioned, and inspectable
  • 🔐 Systems where consent is always provable

🫱🏽‍🫲 Final Words

We’re not here to “own” trust.
We’re here to make it programmable — and make sure it's used responsibly.

Let’s build a world where humans own the keys, the scopes, and the outcomes.

Consent-first. Privacy-native. Agent-ready.

— Team Hussh