Key Concepts
This page is a glossary of the terms used throughout these documents. Each entry gives a short definition and, where there is one, a link to the page that covers it in full.
Agent An Assistant that another Assistant can hand a task to. The Agent works on the task separately, with its own tools, and reports back. See Assistants.
API Key A credential that lets an application or script use the OnVeris API. A User Key acts as one person; a Service Key has only the scopes it is given. See API Keys.
Assistant A ready-made AI helper for a particular job, set up by an administrator: a model with instructions, and optionally tools, Knowledge Bases and Agents. See Assistants.
Backend The build of llama.cpp that runs language models on a Node's hardware — for example, Vulkan, CUDA, ROCm or Metal. See Hardware and GPU.
Connector A tool, or set of tools, the AI can use to work with another system. OnVeris Connectors and MCP Connectors are managed on the Gateway's Connectors page.
Context Window How much text a model can hold at once — its working memory for a request. Measured in tokens. See Context Window.
Conversation A saved thread of messages in OnVeris Chat. Conversations are stored on the Gateway and belong to the person who had them.
Embedding Model A small model that turns text into numbers that can be compared by meaning. Used by Knowledge Bases, AI memory and the embeddings API.
Executor The process a Node runs for one model. Each model you start has its own. See Executors.
Gateway The central OnVeris server. Handles sign-in, routing, tools, Assistants, Knowledge Bases and the administration pages. There is typically one Gateway per installation.
GGUF The file format for AI model weights that OnVeris uses. GGUF models are downloaded from Hugging Face and run by llama.cpp.
Instance One of several copies of the same model running on one Node, each on its own GPU, so more requests are answered at once. See Hardware and GPU.
Knowledge Base A collection of documents the AI can search by meaning, to answer from your own information. See Knowledge Bases.
MCP Server A Model Context Protocol server — a service, often run by another company, that provides tools the AI can call. Added to OnVeris as an MCP Connector.
Model An AI model, such as a chat model or an embedding model, stored as a GGUF file on a Node.
Node An inference server connected to the Gateway. Each Node runs models on its own hardware and answers the requests the Gateway sends it. You can connect several Nodes to one Gateway.
Projector A file that, loaded beside a language model, lets the model read images. See Executor Settings.
Scope
One area of the API, such as chat.completions or knowledge.read. An API Key can use only the scopes it holds. See API Keys.
System Prompt Instructions given to the model before any of the conversation, shaping its role, tone and behaviour. In OnVeris, an Assistant's Instructions are its system prompt.
Token A piece of a word — the unit models read and write. As a rough guide, 1,000 tokens is about 750 words of English.
Tool Something the AI can do besides writing text — for example, look something up, read a file, or fetch data from another system. Tools come from Connectors and from OnVeris itself.
Workspace The folder on a person's computer that a conversation works in, for tools that run on that computer. See Using Tools.