Tarseek API · Pre-production

Embeddings

POSThttps://tarseek.com/v1/embeddings

Body: model, input (string or array of strings), optional dimensions, optional encoding_format (float | base64).

Implemented

The embeddings endpoint uses project API keys, policies and the project's BYOK provider credential. Native OpenAI and compatible OpenAI embeddings deployments are supported. Other provider protocols are explicitly unavailable for this operation.

Send 1–2048 non-empty text inputs, up to 1 MB in total. Token arrays, multimodal inputs, streaming and truncation settings are rejected. Supported dimensions are passed to the provider; Tarseek never silently slices a vector or substitutes another model.

Results preserve input indices and include prompt-token usage, total tokens and cost. Invalid vector counts, non-finite values, inconsistent dimensions or missing usage fail closed. Vectors are not retained by default. OpenAI text-embedding models use their documented tokenizer; other compatible deployments use conservative admission estimates and provider validation.