Glossary

The terms that come up when you integrate an LLM API. Definitions are short on purpose.

Abliterated

A model whose refusal direction has been removed from its weights. The edit is baked into the model, not applied by a system prompt, so it holds across every request.

Uncensored

A model id with the provider's safety refusals relaxed. Softer than 7api: it changes behaviour without editing the weights.

Cached tokens

Prompt tokens served from the provider's prompt cache instead of being recomputed. They bill at a discounted cached rate; repeated system prompts benefit most.

Context window

The maximum tokens a model can hold at once — your input plus its output. Ranges from 128K to about 1.05M in this catalog.

Base URL

The root you point an SDK at. Here it is https://uncensoredai.dev/v1. Swapping it is the whole migration from OpenAI.

Bearer token

The credential in the Authorization header. Here it is your 16-digit access key: Authorization: Bearer sk-ua-…

Streaming / SSE

Server-Sent Events. With stream: true the API emits text/event-stream, one partial delta per data: line, terminated by data: [DONE].

Function calling / tools

Passing a tools array so the model can request a structured call your code executes. Also called tool use. Not every model supports it.

Token

The unit a model reads and writes; roughly a few characters of English. All billing here is per 1,000,000 tokens.

Per-token pricing

Charging by tokens consumed rather than per request, seat or month. Input, cached and output each have their own rate.

Prepaid balance

Credit in USD you spend down as you make requests. When it hits zero, calls fail with 402 insufficient_quota.

KYC

Know Your Customer — identity verification usually required before an account is opened. UncensoredAI does not perform it; there is no account identity to verify.