Glossary

Current BullSequana AI product, tenancy, model access, usage, and operational terminology.

Agentic Friendly

Product and workspaces

TermMeaning
AIProduct domain for models, RAG, assistants, voice, and agents.
DataProduct domain for integration, lakehouse, catalog, data science, orchestration, streaming, and BI.
FoundationProduct domain for shared identity, inference, networking, observability, storage, databases, and delivery services.
BSQAI APIStable platform-facing interfaces used by applications and supported tools.
PortalUser-facing web experience for BullSequana AI capabilities.
OmniChatPortal workspace for chat, files, Agents, Skills, Connectors, and API keys.
StudioPortal workspace for models, model deployment, the data catalog, data ingestion, and compute clusters.
AdministrationPortal workspace for tenant and platform administration. It is shown only to platform administrators, tenant administrators, and members who may read their tenant's usage.
BuildChat mode in which an agent writes and runs code in the code sandbox.
Code sandboxIsolated, per-chat workspace where agents and Skill scripts run code. The site enables it; platform and tenant administrators can turn it off.

Tenancy and identity

TermMeaning
InstallationOne deployed BullSequana AI platform.
TenantCustomer or organization isolated from other customers on an installation. The Portal calls it an organization.
Global OrganizationThe installation's own organization. It owns the platform's models and services, which can be offered to other tenants.
GroupSet of tenant members and service accounts that roles, model allow rules, and BCU caps can target. A group stands for a team, such as a department or working group.
Platform administratorAdministrator with installation-wide and tenant-lifecycle responsibilities, including each tenant's quotas, deployment policy, publishing permission, and BCU allowance.
Tenant administratorAdministrator responsible for identity, groups, membership, service accounts, and delegation inside one tenant.
Service accountNon-human identity for automation in one tenant, with its own client credentials, roles, groups, and owners. See Manage service accounts.
System roleBuilt-in, fixed permission bundle, such as Data Builder or Inference Deployer. See System roles.
Custom roleTenant-defined, additive selection of permissions. See Roles and permissions.
API keyPersonal, tenant-bound credential with an immutable permission scope that never exceeds its owner's current access.
TeamPortal label for the group an API key is bound to when it is created. The key's calls count against that group's BCU cap.

Models and deployment

TermMeaning
Model ownerPerson who deployed a model. The owner calls it without an allow rule while a member of its organization.
Allow ruleCan use rule that lets All Members, one group, or one person in an organization call a model. Rules apply to Portal sessions and API keys alike. See Model access.
Grant ManagerPerson or group that manages a resource's sharing, or a model's allow rules, without owning it.
PublishingA tenant offering one of its language models to every tenant. A platform administrator must allow the tenant to publish, and each receiving tenant sets its own allow rules.
Predictive modelModel that returns predictions from structured inputs rather than generated text, served as a KServe InferenceService.
Deployment policyThe models, serving runtimes, and resource profiles a tenant may deploy, set by platform administrators. See Set what a tenant may deploy.
Resource profilePlatform-defined runtime profile that sets a model deployment's image, resources, and runtime arguments, such as a CPU or a 1-GPU vLLM profile.
Serving runtimeKServe runtime a model is served with, such as vLLM for language models or a predictive-model runtime.
Private MLflowA tenant's own MLflow instance, which holds its experiments, registered models, and the artifacts its model servers read.
GPU allocationThe Guaranteed GPUs and GPU limit of all of a tenant's model deployments together, enforced by its GPU scheduling queue.
Compute quotaThe CPU, memory, and storage cap on a tenant's namespace.

Usage and limits

TermMeaning
BCUUnit in which BullSequana AI measures consumption: model tokens and the capacity that a tenant's running pods hold.
AllowanceA tenant's soft monthly BCU limit, set by platform administrators. Reaching it refuses nothing. See Manage monthly BCU allowances and caps.
CapHard monthly BCU limit on a member or on the group that funds API keys. Reaching it refuses new model requests it covers.

MCP Connectors

TermMeaning
ConnectorModel Context Protocol (MCP) server that a tenant administrator, or a member with mcp.server.create, has made available to an organization.
Managed ServerConnector whose container the platform runs inside the cluster: in its organization's tenant namespace, or in the shared mcp-servers namespace for the Global Organization.
Remote ServerConnector that runs outside the platform, reached at a stored address.
MCP GatewaySingle authenticated entry point that every MCP call passes through.
UpstreamThird-party system a Connector acts against, such as GitHub or Linear.
Credential ModeWhose identity a Connector reaches its Upstream as, chosen when the server is added: the calling user's own, or none.
User CredentialSecret that authorizes one user against one Connector's Upstream. An unauthenticated Connector has none.
LinkingAct of a user supplying their User Credential for a Connector.
Tool AllowlistSubset of a Connector's tools that the organization's members may call, set by whoever manages the Connector.

Historical product names can appear in release notes, repository names, image paths, and deployment identifiers. They do not define the current documentation taxonomy.

On this page