Key Concepts

Understanding core concepts in Bud AI Foundry

Use these definitions to navigate the Bud AI Foundry platform and documentation.

1. Projects

Projects are the primary workspace for teams. Each project contains its own models, deployments, endpoints, and API keys so you can isolate workloads by application or business unit.

2. Models

Models are registered artifacts or managed APIs that power inference. Bud AI Foundry supports cloud providers, Hugging Face repositories, URL-based imports, and disk-mounted checkpoints, all governed under a single catalog.

3. Clusters

Clusters represent the infrastructure where models run. They can be cloud-managed Kubernetes clusters or on-prem hardware pools, and they define the compute profiles available to deployments.

4. Deployments

Deployments are the runtime instances of models attached to clusters. They include configuration details like hardware settings, scaling policies, and safety controls.

5. Endpoints

Endpoints expose deployed models to applications through OpenAI-compatible APIs. They can include routing rules, fallback strategies, and access policies.

6. API keys

API keys authenticate application calls to Bud AI Foundry endpoints. Keys are project-scoped, support rotation, and can be segmented by environment.

7. Observability

Observability covers metrics, logs, and traces for latency, token usage, and error rates so teams can monitor reliability and cost.

8. Evaluations

Evaluations measure quality, latency, and safety using datasets and benchmarks so teams can choose the best model and configuration before promoting to production.

9. Guardrails and Bud Sentinel

Guardrails enforce safety policies during inference, while Bud Sentinel adds evaluation and runtime trust checks to validate models and responses.

10. Playgrounds

Playgrounds provide a safe environment for prompt iteration, model comparison, and quick prototyping before integrating APIs.