Database-first security

Govern memory, retrieval, and tools at the source of truth.

MaluDB treats AI security as a database systems problem: roles, schemas, privileges, Row Level Security, compartments, audit trails, and rebuildable derivations.

Identity
Users, services, agents, tools
Boundary
Compartment-level retrieval and row policies
Audit
Prompt, model, source, tool, and memory events

Compartmentalized memory

Sensitive context can be separated by tenant, product, department, incident, account, project, or regulatory boundary.

Retrieval with policy checks

Search plans apply identity and compartment rules before context is assembled for an application, model, or MCP tool.

Provenance and audit

Memory changes can point back to source packages, extraction jobs, validation decisions, model calls, and human reviewers.

Operational resilience

Preserved sources and derivation records make it possible to rebuild memory indexes after policy, schema, or model changes.

Controls

Security surfaces to document before launch.

This scaffold leaves room for a formal security page with compliance posture, incident response, vulnerability reporting, and deployment architecture.

AreaControlWebsite message
AccessPostgreSQL roles, privileges, and RLSPolicies stay close to data and memory.
AI usagePrompt, context, model, and tool permissionsAgents can only see and do what the DB allows.
TraceabilitySource links and derivation ledgerTeams can explain how memory was created.
IsolationCompartments and deployment boundariesMemory is separated by organizational need.