Open Memory Standards
for AI Agents

MemCube v4.1 and SMRS v1.0 — formal specifications for interoperable, provably safe agent memory governance.

View MemCube v4.1 Spec View SMRS v1.0 Spec

MemCube v4.1 — Memory Entry Schema

A standardized JSON Schema for AI agent memory entries. Designed for interoperability across LangChain, CrewAI, AutoGen, LangGraph, mem0, and any agent framework.

7 Required Fields

Field Type Description
idstringUnique identifier
contentstringMemory content text
typeenumOne of 7 memory types
timestamp_age_daysfloatAge in days since creation
source_trustfloat 0-1Source reliability score
source_conflictfloat 0-1Contradiction level with other sources
downstream_countintNumber of dependent agents

Memory Types

episodic semantic preference tool_state shared_workflow policy identity

Example Entry

{
  "id": "mem_001",
  "content": "User prefers wire transfers under $10,000",
  "type": "preference",
  "timestamp_age_days": 14,
  "source_trust": 0.91,
  "source_conflict": 0.04,
  "downstream_count": 3
}

SMRS v1.0 — Memory Risk Score

A formal risk scoring standard for AI agent memory validation. Computes a 0-100 risk score from 10 components, producing one of four preflight decisions.

Decision Bands

Sgraal returns one of four decisions per preflight call. Concrete numerical band cutpoints are tenant-specific and recalibrated continuously — only the qualitative semantics are part of the public standard.

USE_MEMORY

Safe to proceed

WARN

Proceed with caution

ASK_USER

Human approval needed

BLOCK

Do not proceed

10 Risk Components

s_freshness

Memory age decay

s_drift

Semantic drift

s_provenance

Source reliability

s_propagation

Blast radius

r_recall

Recall accuracy

r_encode

Encoding quality

s_interference

Cross-entry conflict

s_recovery

Recovery capability

r_belief

Model belief alignment

s_relevance

Intent-drift detection

Standard body: Sgraal Governance Working Group · Version: SMRS v1.0

Empirically Validated

Joint benchmark with Grok (xAI)

1.000

F1 Score

Across all 3 corpora

239/239

Corpus Cases

All green on live API

0

False Negatives

No unsafe memory passed through

View corpus on GitHub →

These figures reflect synthetic R12/R14 corpus performance; production calibration is pending paying-customer onboarding.

Compatible With

LangChain CrewAI AutoGen LangGraph mem0 OpenAI Agents SDK
View Integration Docs →
Concept · Coming

For Regulators

Reference implementations for AI accountability.

Regulators need reference implementations and standards specifications to enforce AI accountability. The Sgraal protocol could become an industry standard — we welcome regulator engagement, working group seats, and RFC participation.

science

AI Black Box Recorder reference impl

A reference deployment of the AI decision recorder pattern — deterministic replay, signed evidence, court-admissible artefacts. Designed to be cited in regulatory guidance rather than to compete with regulator-developed alternatives.

description

Sgraal Standard specification

Formal specification document for the memory governance protocol — published openly, open to comment, structured for standards-body submission. The draft is in active development; we welcome regulator review.

forum

Working group seat · RFC participation

We are interested in participating in working groups and responding to RFCs related to AI memory governance, EU AI Act enforcement, and emerging accountability standards.

How to engage with us as a regulator

If you represent a regulator, supervisory authority, or standards body and want to discuss AI memory governance, the reference implementations above, or the draft specification — email below. We respond.

Email the regulator desk →

No demo calls or sales follow-ups. Regulator engagement, not vendor pitching.

Start validating memory today

No signup needed. Use the demo key to try the full 87-module safety pipeline.

See pricing

Demo key: sg_demo_playground — no signup needed