Core concepts
The shared product and responsibility model used throughout BullSequana AI documentation.
Use these terms consistently when reading or writing BullSequana AI documentation.
Product domains
| Domain | Responsibility |
|---|---|
| AI | User-facing AI experiences and reusable AI capabilities such as models, RAG, voice, and agents. |
| Data | Data integration, engineering, catalog, data science, orchestration, streaming, and analytics capabilities. |
| Foundation | Shared platform services for identity, inference, networking, observability, storage, databases, and delivery tooling. |
These domains describe the current product. Historical names can still appear in release notes or technical identifiers, but they are not used as product areas.
Resource hierarchy
BullSequana AI provides multi-tenant isolation and team-scoped access as part of the product authorization model.
Installation
├── Global Organization
└── Tenant / customer organization
└── Team (group)
├── Users and service accounts
├── Applications and agents
├── Data and knowledge
└── Models and workloads- An installation is one deployed BullSequana AI platform.
- A tenant represents a customer or organization. The Portal calls it an organization, and users with several memberships choose the active one. BullSequana AI isolates its resources and administration from other tenants.
- The Global Organization is the installation's own organization. It is a usable default tenant, and it owns the platform-provided models and services that can be offered to other tenants. Platform administrators administer it.
- A team is a department or working group inside one tenant. In the Portal, teams are tenant Groups, and an API key's Team is one of them. Teams provide delegated access without weakening the tenant boundary.
- Members and service accounts receive roles and groups in a tenant. A person can be a member of several tenants in the Portal and the BSQAI API. The Developer Workspace serves only users who belong to exactly one tenant, the Global Organization included, so a person who works with data for several tenants has a separate user in each; see Manage members. A service account is a non-human identity for automation in one tenant; it belongs to the tenant, not to the person who created it.
- The product scopes access, applications, data, models, and workloads through this hierarchy.
Models and access
- Models are owned. A model a tenant deploys belongs to that tenant, and the person who deployed it is its owner. The Global Organization owns the platform's models.
- Calling a model needs an allow rule. A Can use rule lets all members, a group, or one person in an organization call a model. Tenant administrators, the model's owner, and its Grant Managers manage the rules. The owner calls the model without a rule. For anyone else, a model without a rule cannot be called, and model pickers and the BSQAI API model list leave it out.
- Models can be offered to other tenants. When platform administrators allow it, a tenant can publish one of its language models to all tenants. Each receiving tenant still decides with its own allow rules which of its members may use it.
See Model access.
Consumption and limits
- BCU is the unit BullSequana AI measures consumption in. Every model call is recorded with its user and, for an API key, the key and its funding team.
- Platform administrators set each tenant's monthly BCU allowance. It is soft: service continues when it is reached, and the overage is shown.
- Tenant administrators set monthly BCU caps on members and on the teams that fund API keys. A reached cap refuses new model requests.
- Platform administrators also set each tenant's quotas: the CPU, memory, and storage its namespace may use, and the GPUs its models are guaranteed and capped at. A tenant's deployment policy decides which models, serving runtimes, and resource profiles it may deploy.
See Manage monthly BCU allowances and caps, Set a tenant's quotas, and Set what a tenant may deploy.
Runtime isolation
Each customer tenant runs in its own Kubernetes namespace, tenant-<slug>, with a resource quota and a limit range. When network policies are enabled, as they are by default, the namespace starts with a default-deny network policy. A tenant also gets its own identity organization and, when selected, its own object storage bucket, application database, vector database, private MLflow, and GPU queue for model serving. The Global Organization's models run in the platform's own serving namespace. See Tenant operator.
Responsibility boundaries
- Deployers prepare infrastructure and install the platform.
- Platform engineers keep the installation healthy and recoverable.
- Platform administrators manage installation-wide settings, the Global Organization's models, and tenant lifecycle, quotas, deployment policies, and BCU allowances.
- Tenant administrators manage identity, groups, membership, service accounts, model access, and BCU caps inside one tenant.
- Developers and data practitioners build applications, models, and pipelines without operating the underlying platform.
- End users consume governed AI and data experiences.
- Security and compliance owners approve data and use-case access and collect evidence.