Capability domains
The main capability areas covered by AI and the components behind each.
| Capability | Main components | Purpose |
|---|
| Web experience | AI Web Portal | OmniChat for chat, Build, files, Agents, Skills, Connectors, Deep Research and voice; Studio for models, data and compute; and administration |
| Backend services | BSQAI API, Temporal | Tenant-aware application APIs and durable workflows |
| Developer compatibility | BSQAI API | Chat Completions, Responses, Models, Embeddings and Rerank APIs under /v1, and the Anthropic Messages API under /anthropic, for applications, SDKs and coding tools |
| Application identity | BSQAI API, Keycloak | API keys with a permission scope, and service accounts that give applications their own identity |
| Capability | Main components | Purpose |
|---|
| Model registry | MLflow | Track experiments, models, versions, and GenAI traces, in the installation's MLflow or an organization's private MLflow |
| Model delivery | Model Installer | Import, register, and deploy model artifacts |
| Model serving | KServe, vLLM, MLServer, NVIDIA Triton, KAI Scheduler | Deploy, schedule, scale, and execute LLMs and predictive models, within each tenant's GPU guarantee and cap |
| Model access | BSQAI API, OpenFGA | Publish models to other organizations and grant their use with allow rules |
| Usage and budgets | BSQAI API usage records, Analytics, Budgets | Attribute BCU consumption to users, API keys, applications, models and groups, and cap it monthly |
| Capability | Main components | Purpose |
|---|
| Model routing | BSQAI API | Send each model request to KServe or to a configured external model server |
| MCP operations | MCP Gateway, MCP Lifecycle Operator | Manage, authorize, and call managed or remote MCP Servers |
| Web search | BSQAI API | Configure tenant providers and supply the web_search tool |
| Speech | Speaches | Batch or real-time transcription with FasterWhisper, and streamed speech generation with Piper and Kokoro voices |
| Code execution | Agent Sandbox, Kata Containers | Run the agent's commands and skill scripts in isolated sandboxes, where the site enables them |
| Capability | Main components | Purpose |
|---|
| Document conversion | Docling | Convert documents into structured content |
| Vector search | Milvus | Store and query embeddings for tenant-authorized retrieval |
| Workflow orchestration | Temporal | Run durable ingestion, reindex, and document-processing operations |
| Capability | Main components | Purpose |
|---|
| GenAI observability | OpenTelemetry, MLflow | Trace inference, retrieval, tool, storage, and workflow operations |
| Telemetry routing | Grafana Alloy | Route application and inference signals to Foundation backends |