Skip to main content

​
Endpoint Model

Each integration request targets a project endpoint exposed by a deployed model. In practice, requests combine:
  • Base URL from the deployment/project route
  • API path such as /v1/chat/completions or /v1/embeddings
  • Model value expected by the selected endpoint

​
Authentication and Scope

Bud uses bearer token authentication with project-scoped API keys. Required header:
  • Authorization: Bearer <API_KEY>
Project scoping ensures:
  • Team-level separation
  • Usage attribution per project
  • Easier key rotation and revocation

​
Request Patterns

Most endpoints follow OpenAI-style JSON requests. Some endpoints (for example, audio transcription) require multipart form uploads.
Endpoint TypeCommon PathBody Style
Chat/v1/chat/completionsJSON
Completions/v1/completionsJSON
Embeddings/v1/embeddingsJSON
Classify/v1/classifyJSON
Image Generation/v1/images/generationsJSON
Audio Transcription/v1/audio/transcriptionsMultipart form
Text-to-Speech/v1/audio/speechJSON
Documents/v1/documentsJSON

​
Reliability Concepts

Timeouts Define client timeouts that match endpoint workload characteristics. Retries Use exponential backoff for transient errors (429, 5xx), and avoid retrying invalid requests (400, 401). Idempotency Use idempotency keys when supported by your integration pattern to protect against duplicate submissions.

​
Observability Concepts

Track these signals from day one:
  • Request latency percentiles
  • Error rate by status code
  • Request and token volume
  • Endpoint-level throughput
This helps correlate model behavior, scaling limits, and user experience.