Skip to content

What is a2p?

a2p (Agent 2 Profile) is an open protocol that enables user-sovereign AI profiles. It gives users ownership and control of their AI interaction data across all AI agents.

The Core Idea

Think of a2p as a digital passport for AI interactions:

Passport a2p Profile
Owned by you Owned by you
Contains your identity Contains your preferences
Portable across countries Portable across AI agents
You control what to share You control what to share

Why a2p Exists

The Current Problem

When you use AI assistants today:

  1. Your data is siloed — Each AI has its own copy of your preferences
  2. You don't own it — The AI company owns your interaction history
  3. It's not portable — Switch to a new AI? Start from scratch
  4. Privacy is unclear — Hard to know what's stored or delete it

The a2p Solution

With a2p:

  1. You own your profile — Cryptographic ownership via DIDs
  2. It's portable — Works with any a2p-compatible agent
  3. You control access — Granular consent policies
  4. Compliance built-in — GDPR and AI Act requirements native

Key Concepts

Profiles

A profile contains:

  • Identity — Display name, pronouns (optional)
  • Preferences — Language, timezone, communication style
  • Interests — Topics, hobbies, professional context
  • Memories — Things agents have learned about you
{
  "id": "did:a2p:user:local:alice",
  "profileType": "human",
  "identity": {
    "displayName": "Alice",
    "pronouns": "she/her"
  },
  "common": {
    "preferences": {
      "language": "en-US",
      "communication": { "style": "concise" }
    }
  },
  "memories": {
    "a2p:professional": {
      "occupation": "Software Engineer"
    }
  }
}

You define rules for who can access what:

{
  "agentPattern": "did:a2p:agent:local:my-assistant",
  "allow": ["a2p:preferences.*", "a2p:interests.*"],
  "deny": ["a2p:health.*"],
  "permissions": ["read_scoped", "propose"]
}

Memory Proposals

Agents can't just write to your profile. They propose memories, and you approve:

  1. Agent learns something: "User prefers TypeScript"
  2. Agent proposes this as a memory
  3. You review the proposal
  4. You approve, edit, or reject

How It Works

sequenceDiagram
    participant User
    participant Agent
    participant Profile

    Agent->>Profile: Request access (with purpose)
    Profile->>Profile: Check consent policies
    Profile-->>Agent: Grant/deny access

    Agent->>Profile: Read allowed data
    Profile-->>Agent: Return filtered profile

    Agent->>Profile: Propose memory
    Profile->>User: Review proposal
    User->>Profile: Approve/reject

Who Is It For?

For Users

  • Control your AI interaction data
  • Portable preferences across agents
  • Easy GDPR rights exercise

For AI Developers

  • Get user context without cold start
  • Built-in compliance
  • Framework adapters ready

For Enterprises

  • Standardized AI data governance
  • Reduced compliance overhead
  • Vendor portability

Open & EU-Made

  • License: EUPL-1.2 (EU open source license)
  • Location: Developed in Madrid, Spain
  • Values: Privacy, sovereignty, interoperability

Next Steps