Sothe API

What Sothe does

Sothe puts 54 models from 14 providers behind one endpoint, one key and one balance. You top up credits once and spend them on any model, at per-token prices published in the model list — no subscription, no minimum commitment, no separate account per provider.

Every request is billed on the tokens it actually used, logged with its cost, and capped by whatever limits you set on the key, the workspace and your guardrails.

OpenAI compatibility

The API implements OpenAI's Chat Completions schema. If your code already calls OpenAI, three changes are enough: the base URL, the key, and the model name.

1- base_url="https://api.openai.com/v1"2+ base_url="https://api.sothe.net/v1"

Streaming, tool calling and multimodal input all work through the same fields you already use. A handful of OpenAI parameters are accepted but not yet acted on — they are listed plainly under Chat completions so nothing fails silently.

How a request works

1. Your key is checked, along with its per-minute allowance.
2. The request is translated to the model provider's own format.
3. The worst case the request could cost is reserved against your balance — this is why an underfunded request is refused before any tokens are generated.
4. The model runs, streaming back to you if you asked for it.
5. The real usage is charged, the reservation released, and a row written to Logs.

Where to start

Go to Quickstart for a working request in Python, TypeScript or curl. If you are moving an existing integration over, Chat completions and Errors are the two pages worth reading in full.