API and web terms
API
API
An Application Programming Interface is a defined way for software systems to communicate. Your application sends a request; the API returns a response.
Base URL
Base URL
The shared beginning of API addresses. LLM.kiwi uses
https://api.llm.kiwi/v1.Endpoint
Endpoint
A specific API function and path. For example,
POST /v1/chat/completions creates a chat completion, while GET /v1/models lists models.HTTP method
HTTP method
The action associated with a request.
GET retrieves information. POST sends data for processing.Header
Header
Metadata sent with a request. The
Authorization header carries the API key, and Content-Type tells the API that the request body is JSON.Request body
Request body
Data sent with a request. For a chat completion, the JSON body contains fields such as
model and messages.JSON
JSON
JavaScript Object Notation is a text format for structured data. JSON uses double-quoted keys and values, objects in braces, and arrays in brackets.
HTTP status code
HTTP status code
A number describing the result.
200 means success, 400 means the request is invalid, 401 means authentication failed, and 429 means a rate limit was reached.Timeout
Timeout
The maximum time a client waits for a request. A timeout prevents an application from waiting forever when a network or service problem occurs.
Authentication terms
API key
API key
A secret value identifying authorized API access. LLM.kiwi keys begin with
sk_kiwi_. Treat a key like a password and keep it on a trusted server.Bearer token
Bearer token
The authentication format used in the request header:
Authorization: Bearer YOUR_API_KEY. The space after Bearer is required.Environment variable
Environment variable
A value supplied to a program by its operating environment instead of being written directly in source code. It is the normal place to load an API key, though deployment access controls still matter.
Revoke
Revoke
Disable an API key so it can no longer be used. Revoke a key immediately if it was exposed.
AI terms
LLM
LLM
A large language model is an AI model trained to process and generate language. It predicts output based on the instructions and context in a request.
Model
Model
The AI system selected for a request. LLM.kiwi documents
auto for automatic selection and hrllm for Croatian-focused generation.Prompt
Prompt
The input and instructions given to a model. In chat completions, this usually means the complete
messages array, not only the newest user message.System message
System message
A message with role
system that describes the assistant’s behavior or task. It guides output but does not replace application-level security or validation.Token
Token
A unit a model uses to process text. A token may be a whole word, part of a word, punctuation, or whitespace. The count varies by language, text, and tokenizer.
Context window
Context window
The maximum token capacity available for input and output in a request. When a request is too large, your application must shorten, remove, split, or summarize content.
Temperature
Temperature
A sampling setting that can affect output variation when supported by the selected model. Exact behavior and allowed values can differ between models.
Completion
Completion
Text generated by the model in response to the request. In a chat completion, the main result is usually at
choices[0].message.content.Finish reason
Finish reason
A response field describing why generation ended.
stop commonly indicates a normal end; length commonly means an output or context limit was reached.Streaming
Streaming
Receiving a response incrementally as it is generated instead of waiting for one complete response.
Hallucination
Hallucination
A confident-looking but incorrect or unsupported model output. Verify important facts and validate outputs before acting on them.
Application terms
SDK
SDK
A software package that wraps HTTP requests in language-specific functions and types. LLM.kiwi can be used with compatible OpenAI clients by setting a custom base URL.
Rate limit
Rate limit
A restriction on request frequency or capacity. When reached, an API commonly responds with status
429.Exponential backoff
Exponential backoff
A retry strategy that increases the wait after each temporary failure, for example one second, then two, then four. A small random delay helps prevent many clients from retrying together.
Stateless API
Stateless API
An API that does not remember prior requests automatically. For a continuing conversation, your application sends the relevant message history each time.
Server-side code
Server-side code
Code that runs in infrastructure you control and can safely access secrets. Browser JavaScript is client-side and must not contain private API keys.
Learn these concepts in context
The core concepts guide connects these terms into one complete request flow.