Core Concepts
Workspaces & Tenancy
Knowledge boundary isolation, personal namespaces, and organization tenancy.
Workspaces form the fundamental scoping boundary in Omni. A workspace isolates an agent's memory, uploaded documents, conversation threads, and configuration.
Storage Namespaces
All workspace data is partitioned into secure namespaces managed server-side:
| Namespace Type | Storage Pattern | Access Model | Status |
|---|---|---|---|
| Personal | projects/<userId>/workspaces/<workspaceId>/ | Bound strictly to the authenticated user identity | Live |
| Organization | orgs/<orgId>/workspaces/<workspaceId>/ | Shared enterprise tenant with role-based member access | Preview |
[!IMPORTANT] No Raw Storage Access: Clients and developers never interact directly with underlying object storage buckets. All storage paths and access policies are verified and synthesized server-side by
edge.omnistatic.com.
Workspace Structure
Within each workspace, Omni maintains distinct knowledge domains:
workspaces/<workspaceId>/
├── configuration # Agent persona and workspace metadata
├── profile & bio # Biographical facts automatically injected into prompts
├── documents/ # Structured knowledge documents and markdown assets
└── memory/ # Persistent vector-indexed semantic memoriesMulti-Tenant Organization Model (Preview)
In enterprise deployments, organizations provide shared agent collaboration:
- Tenant Ownership: The organization creator acts as the root administrative principal.
- Membership Verification: The edge control plane verifies active membership on every request before admitting access.
- Resource Binding: API keys can be explicitly bound to an
orgId, restricting token operations to that organization's perimeter.