Key Concepts

A glossary of terms used throughout these documents.

Gateway The central OptimaGPT server. Handles authentication, routing, the admin interface, and the chat UI. There is typically one Gateway per installation.

Node An inference server connected to the Gateway. Each Node discovers GGUF model files on its local disk and handles inference requests sent to it by the Gateway. You can connect multiple Nodes to a single Gateway.

Executor The inference process that a Node runs for a particular model. When a model is enabled on a Node, the Node launches a dedicated executor for it. Each enabled model runs in its own executor process.

Model A GGUF-format AI model file. Models are placed in a specific directory on the Node's host machine and are discovered automatically. Each enabled model gets its own executor.

Conversation A saved thread of messages between a user and a model in OptimaChat. Conversations are stored by the Gateway and are accessible from the conversation sidebar.

API Key A credential that allows external applications or scripts to authenticate with the Gateway API. API keys are created and managed in the Gateway admin interface.

MCP Server A Model Context Protocol server — an external service that provides tools (such as web search, database access, or file operations) that AI models can call during a conversation. MCP servers are configured in the Gateway and made available to users automatically.

GGUF The file format used for AI model weights. GGUF files are downloaded from model repositories and placed in the Node's model directory. OptimaGPT uses llama.cpp to run them.

System Prompt An instruction given to the AI model at the start of a conversation, before any user messages. It shapes the model's tone, role, and behaviour. Users can set a system prompt from within OptimaChat.