Memory Passport

Cross-agent memory that users can actually control.

Users approve categories, manage grants, answer pending conflicts, correct stale memories, and revoke access without every AI app rebuilding the same permission center.

Signal received from a verified agent

Study Buddy can read goals and learning preferences

received

01

Career Coach can read skills and project history

received

02

Personal conflicts wait for user confirmation

received

03

Revoked grants stop future access without deleting history

received

04
Building compact context from approved Passport categories…

Shared memory problem

Why this needs memory infrastructure.

The hard part is not remembering text. It is deciding what is durable, what is stale, who controls it, and what the next agent should trust.

Shared memory becomes risky if users cannot see what each agent knows.

Agent-level consent is too broad unless categories are visible and revocable.

Conflicting personal facts should not be silently resolved by an app developer.

Users need one place to correct, remove, and inspect memories across trusted agents.

What you get

Clear context for the next model call.

Visible grants

Users see which agents can read which categories before and after approval.

Agent identity

Tenants register public agent identities before asking for Passport access.

Pending questions

Personal conflicts route to the user instead of being guessed by the system.

Revocation

Users can revoke access later while historical memories remain separately manageable.

Source status

The UI can show active, expired, revoked, disputed, or corrected source state.

Cross-agent context

Approved memory can travel across trusted AI apps without losing user control.

Integration

Use the SDK first. Add governance when needed.

The same API supports solo apps and production teams. Start with a stable user ID, then add source metadata, service writers, or Memory Passport when your product needs them.

Working pattern
from memoryos.universal import UniversalMemory

consent_url = UniversalMemory.consent_url(
    agent_id="your_global_agent_id",
    redirect_uri="https://yourapp.com/memoryos/callback",
    categories=["preference", "fact", "goal"],
    state="your_user_session_id",
)

# Redirect the user to consent_url.
# They decide what this agent can access.

Make shared memory useful without making it hidden.

Register an agent, generate a consent URL, and let users choose what your AI can remember across products.