Tarseek API · Pre-production
Introduction
Tarseek provides one unified API for accessing AI models with provider keys owned by your project. You send an explicit model ID, and Tarseek uses that model's designated primary deployment with the matching project BYOK credential.
Customer architecture
The only objects you need in order to call the API are:
- Developer — the organization that created the model (for example OpenAI, Anthropic, Google, or Mistral).
- Canonical model — the model itself, identified by a stable public ID such as
openai/gpt-5.4. - Provider — the infrastructure or API that actually serves inference.
- Deployment — a specific offering of that model on a provider, with its own pricing, region, and capabilities.
The relationship is: developer → model → deployment → provider.
Implemented
Every request names a model. Tarseek sends it to that model's primary deployment with your project's provider key. If that deployment is unavailable, the request fails.
What you build against
Production Tarseek is an OpenAI-compatible HTTP API. The primary inference surface is POST https://tarseek.com/v1/chat/completions. Model discovery uses both the OpenAI-shaped /models list and the richer Tarseek catalog.
How to read these docs
Each page explains how to call that part of the API. Follow the request shape on the page.