Definitions of terms used in this documentation, written for mixed technical and non-technical audiences.
Agent — AI that pursues a goal by choosing its own steps and using tools, rather than following a fixed script. See Agentic Workflows.
Artifact — A standalone deliverable produced in chat (a document, code file, HTML page) shown in its own panel. See Chat Overview.
Benchmark / Arena — Comparing models side-by-side, or community/model scoring. See Evaluations.
Chunk — A small piece a document is split into for indexing. Chunk size affects retrieval quality. See Knowledge Bases & RAG.
Citations — Numbered source references shown under an answer.
Confidence score — A measure of how certain an extraction or classification is, used to route uncertain cases to human review.
Credit — The internal unit that meters platform usage against your licence. See Usage & Credits.
Deep Agent — A workflow node where an AI plans and acts with tools. See Workflow Nodes.
Embedding — A numerical representation of text that lets the platform find meaning-wise similar passages. The engine behind knowledge-base search.
Enterprise — The commercial licence tier with the full feature set. See Pricing & Licensing.
Fine-tuning — Adapting a model on your own examples so it specialises in your tasks. See Fine-Tuning.
Grounding — Basing answers on supplied sources rather than the model's general knowledge.
Guardrails — Policy rules that check AI inputs/outputs and block violations. See Guardrails.
Hallucination — Confidently wrong content generated when the model lacks the facts. Best reduced with knowledge grounding and citations.
Image-to-video — Animating a still image into a short video. See Video Generation.
Knowledge base — A collection of your documents that AI answers can draw from. See Knowledge Bases & RAG.
LLM (Large Language Model) — The AI that powers conversations and document tasks.
MCP (Model Context Protocol) — The standard that lets the platform plug in external tool servers. See Tools (Admin).
Node — A building block on the workflow canvas. See Workflow Nodes.
OCR (Optical Character Recognition) — Reading text from scanned documents and images.
On-premises — Running the platform on your own infrastructure — the core of the IntraLLM deployment model. See Why IntraLLM.
Permission — A group-level grant controlling feature access. See Roles & Permissions.
Prompt — The instructions and question given to the model. See Prompt Library.
RAG (Retrieval-Augmented Generation) — Finding relevant passages from your documents and giving them to the model to answer from. How knowledge bases work.
Reasoning model — A model that shows intermediate thinking before answering — better for complex problems, slower and costlier.
Role — An account's platform level (Admin, Power User, User, Pending). See Roles & Permissions.
Seed — A number controlling generation randomness — fixed for reproducible results. See Model Parameters.
Skill — A reusable, packaged capability used in workflows and assistants. See Skills.
SSO (Single Sign-On) — Signing in with corporate credentials through your identity provider.
System prompt — Standing instructions that define an assistant's role and rules. Set per template or personally in General settings.
Temperature — Creativity control on model output. See Model Parameters.
Template (Work Station) — A ready-made assistant with instructions, capabilities and knowledge pre-configured. See Using Templates.
Token — The unit AI counts text in (~0.75 words). Drives usage and cost measures.
Tool — A connection that lets AI act on external systems. See Tools.
Valve — In-product name for a model parameter. See Chat Controls.
VLM (Vision Language Model) — An AI that understands images — powers document image descriptions and vision chat.
Web search — Live internet retrieval during a conversation. See Chat Basics.
Workflow — A visual multi-step process on a canvas. See Workflow Builder.