API overview
Choose the protocol your application already speaks.
Updated October 31, 2026
Public base URLs
OpenAI SDK clients use https://api.notholy.com/v1. Anthropic SDK clients use https://api.notholy.com because the SDK adds /v1/messages. All examples use placeholder customer credentials and the public alias notholy-pulse.
Native inference first
POST /v1/chat/completions, POST /v1/responses and POST /v1/messages accept direct inference requests. POST /v1/completions also supports legacy text completions. A prompt array requests independent generations with ordered, zero-based choices and aggregate actual usage. Before the first dispatch, the service reserves one aggregate worst-case hold and one concurrency slot. Legacy completion requests are not implicitly retried. You do not need to create an internal project, conversation or run first.
Schema source
Request and response field documentation is generated from the same versioned contracts that validate the API. Download /v1/openapi.json for the complete machine-readable surface. Additional refinement rules and unsupported features are described alongside the schemas.
Durable first-party workflows
The versioned runs/sessions APIs remain available for CLI integrations needing explicit approvals, artifacts and replay. They are separate from native inference protocols.