AI Core Concepts
LLM Provider
An LLM provider is the external large language model service — such as OpenAI, Anthropic, or a self-hosted engine — that AI Gateway connects to and routes requests toward. Every provider is reached through the same single, OpenAI-compatible entry point. See LLM Providers and Connections.
Token
A token is the unit of text an LLM reads and generates — roughly a word or part of a word. Apinizer measures rate limits, quotas, and cost per token, tracking input, output, and cached tokens separately. See Token Quotas and Rate Limiting.
Embedding
An embedding is a numeric vector that represents the meaning of a piece of text, so that texts with similar meaning end up close to each other numerically. Embeddings are the foundation behind RAG, similarity-based caching, and off-topic detection. See Vector Database.
Vector Database
A vector database stores embeddings and can quickly find the ones most similar to a given query. Apinizer connects to a vector database to power retrieval for RAG and vector-based cache matching. See Vector Database.
Retrieval-Augmented Generation (RAG)
RAG retrieves the most relevant passages from a knowledge base and adds them to a request before it reaches the model, so the response is grounded in your own data instead of relying solely on what the model already knows. See Retrieval-Augmented Generation (RAG).
Knowledge Base
A knowledge base is your own collection of documents — uploaded, automatically split into passages, and indexed — that RAG retrieves from. See Knowledge Bases.
Semantic Cache
The semantic cache reuses a previously generated response for an identical or highly similar request, reducing both latency and cost. See Semantic Cache.
Guardrail
A guardrail is a protective check applied to a request or response — such as masking personal data, blocking a prompt-injection attempt, or catching an oversized request — before it reaches the model or the client. See Advanced Guardrails.
Model Context Protocol (MCP)
MCP is a standard protocol that lets an LLM discover and call external tools and data sources. Apinizer can act as an MCP server, exposing its own tools, or as an MCP client, calling external MCP servers as part of a request. See MCP Gateway.
Agent-to-Agent (A2A)
A2A is a standard protocol for one AI agent to hand off a task to another agent and exchange results. Apinizer can receive tasks as an agent or send tasks to external agents as part of a request. See A2A Gateway.
Failover
Failover automatically routes a request to the next provider or model in a configured chain when the current one fails or is unavailable, keeping the request from failing outright. See Routing and Failover.