Multi-Tenancy

Every Cortex entity is scoped to an app (logical application boundary) and optionally to a tenant (user/organization boundary). This scoping is enforced at the store layer — cross-tenant access is structurally impossible.

Context injection01

Scope identifiers are injected into the Go context using helper functions from the root cortex package:

import "github.com/xraph/cortex"

ctx = cortex.WithTenant(ctx, "org-123")
ctx = cortex.WithApp(ctx, "myapp")

Extraction02

Retrieve scope values from any context:

tenantID := cortex.TenantFromContext(ctx) // "org-123"
appID    := cortex.AppFromContext(ctx)    // "myapp"

Both functions return an empty string if no value is set.

AppID vs TenantID03

Store enforcement04

The PostgreSQL store enforces app scoping on every query:

• Agent, Skill, Trait, Behavior, Persona — all GetByName and List queries filter by app_id

• Run, Memory, Checkpoint — queries filter by tenant_id when provided

• Cross-scope access — returns ErrNotFound even if the entity exists under a different app/tenant

API integration05

When using the Forge extension, the API layer extracts the app ID from the request context (typically set by Forge middleware) and passes it to all engine operations. This means:

• Each API request is automatically scoped to the caller's app

• Agents in app A cannot see or modify agents in app B

• Run history is isolated per tenant within each app

Example: multi-tenant setup06

// Tenant 1 creates an agent
ctx1 := cortex.WithApp(context.Background(), "myapp")
ctx1 = cortex.WithTenant(ctx1, "tenant-1")
eng.CreateAgent(ctx1, agent1Config)

// Tenant 2 creates a different agent
ctx2 := cortex.WithApp(context.Background(), "myapp")
ctx2 = cortex.WithTenant(ctx2, "tenant-2")
eng.CreateAgent(ctx2, agent2Config)

// Tenant 1 cannot see tenant 2's runs
runs, _ := eng.ListRuns(ctx1, &run.ListFilter{TenantID: "tenant-2"})
// runs is empty — tenant isolation enforced