Inference
Discover models and use one canonical API for language, document, speech, and voice inference.
Manage API keys, model access, and organization limits in the console or through the Control API. Send model requests to the separate Inference API.
Tarka exposes model discovery, Chat Completions, Responses, OCR, transcription, translation, speech, and voice cloning at canonical/v1. Native gRPC exposes the same operations. The old/v2 REST prefix is a deprecated alias, not a newer API. The machine-readable OpenAPI 3.1 contractsare published in the public, contract-only API repository.
Choose an integration path
Start with access and authentication, then continue to the modality your application needs.
API access
Choose the public or Agent Host endpoint, configure credentials, discover models, and handle errors.
Language models
Create Chat Completions or Responses and consume token streams with existing OpenAI clients.
Utility models
Use OCR, transcription, translation, speech, and consent-backed custom voices.
Management and runtime traffic are separate
| Surface | Base URL | Use |
|---|---|---|
| Canonical REST API | https://tarka.rest/v1 | OpenAI-compatible language inference plus utility-model routes. |
| Native gRPC API | grpc.tarka.rest:443 | Typed versions of every inference operation. |
| Agent Host inference | http://tarka/v1 | Host-local model requests from Tarka Agent Hosts. |
| Deprecated REST alias | https://tarka.rest/v2 | Temporary compatibility alias for existing integrations. |
| Control API | https://tarka.rest/control/v1 | API keys, access, limits, and provisioned resources. |
| Console | https://tarkahq.com/console | The web interface for the same management resources. |