Skip to main content
TealMemory is a governance module that controls what AI agents can write to and read from memory. It enforces scope boundaries, data classification, and content policies across 5 memory scopes and 4 classification levels.

Overview

Agentic AI systems increasingly use persistent memory — session state, user preferences, shared knowledge bases. Without governance, agents can leak secrets into memory, read across scope boundaries, or store sensitive data in inappropriate classifications. TealMemory introduces 5 new decision actions specifically for memory governance:

5 Memory Scopes

Session, agent, user, shared, and global

4 Classifications

Public, internal, confidential, and restricted

Quick Start


Memory Scopes

TealMemory defines 5 hierarchical scopes:

Scope Hierarchy

Writes to broader scopes require higher classification clearance:
Writing to global scope with internal classification triggers MEMORY_SCOPE_VIOLATION.

Data Classifications

Every memory entry has a classification level:

Classification Enforcement


Policy Configuration

Configure TealMemory through the memory key in your policy:

Write Governance

TealMemory evaluates every memory write against scope, classification, and content policies.

ALLOW_WRITE

Content passes all checks — write proceeds.

DENY_WRITE

Content violates a policy — write is blocked.

REDACT_AND_WRITE

Sensitive content is redacted before writing.

STORE_SUMMARY_ONLY

Large or sensitive content is summarized before storage.

Read Governance

TealMemory also governs memory reads, enforcing scope and classification boundaries.

DENY_READ

Agent attempts to read memory outside its scope or classification clearance.

Reason Codes


Event Types


Python