Skip to main content
The Pearl Inference API exposes multiple providers behind one model catalog and one chat endpoint. Model ids are namespaced as provider/upstream-model-id; unprefixed ids remain Dolphin for backward compatibility.

​
Base URL

​
Endpoints

Chat Completion

POST /v1/inference/chat — unified SSE chat

List Models

GET /v1/inference/models — merged catalog

Providers

GET /v1/inference/providers — capabilities and settlement

Media

Durable images, video, music, jobs, and library

List Templates

Dolphin-only templates

​
Payment kinds

  • none: call immediately; Dolphin is the current free provider.
  • x402: call once for a validated 402 offer, sign it, and retry with PAYMENT-SIGNATURE.
  • prepaid: fund the provider balance using its registry topup block.
  • platform: Pearl collects a fixed configured charge and pays the async upstream using a platform credential; failed/canceled jobs are refunded.
Payment values are additive. Clients should skip unknown values.

​
Durable media

Generated artifacts are captured into Pearl object storage before success is reported. Returned url values are durable public capability URLs suitable for agent memory, later downloads, or chaining into another media request. Library listing and deletion require a short-lived signature from the derived buyer key.
Always fetch the current model/provider catalogs. Provider availability, pricing, modality, and payment mechanics are registry data rather than client constants.