Cortex Memory
Persistent memory for agents
Cortex gives your agents durable, searchable memory with just six methods. It enforces scope for multi-tenancy, supports layered memory, and auto-injects relevant context before execution.
6
Simple methods
Scope
Enforced multi-tenancy
Layers
Company / team / personal
Auto-context
Injected into every run
Architecture
Layers and scope
01
Company Layer
Shared knowledge across your organization (e.g. policies, FAQs). No scope required.
02
Team Layer
Shared memory scoped to a team (e.g. team notes, playbooks). Enforce with scope fields like team_id.
03
Personal Layer
User-scoped memory (e.g. preferences, history). Enforce with scope fields like user_id.
In practice
Auto-context + search
Cortex retrieves recent history and top semantic matches before your agent runs and exposes the formatted results via cortex.context. Scope fields (like customer_id) are enforced automatically on every operation.
cortex query
Cortex Context
Scope: customer_id = “alice”
Retrieving history (limit=10)...
Semantic search (limit=5)...
✓ Injected into cortex.context
“Customer asked about billing on Jan 15th. They are on Premium ($99/mo) and mentioned upgrading as their team grows.”
Methods: store | search | get | update | delete | count