Glossary

A set of connected nodes that runs as a unit. The first node is a trigger; the rest transform or act on the data passed along.

One step in a workflow. It receives items, performs an operation, and outputs items. Nodes are either service-specific (Slack, Postgres) or general (HTTP Request, Code, If).

The first node, which decides when a workflow runs — on a schedule, on an incoming webhook, on an event in a connected service, or manually.

One unit of data moving between nodes: a JSON object, optionally with binary data attached. Nodes receiving many items usually run once per item.

One run of a workflow from start to finish. Executions are the unit plans are billed in, and each is stored with the data that passed through every node.

A stored, encrypted secret for a service — API key, OAuth token, database password — kept separately from the workflows that use it.

A snippet in double curly braces that pulls a value from elsewhere in the run, such as {{ $json.email }}, usable anywhere a field accepts input.

The editor surface where you add and connect nodes.

A group of nodes that work together: a root node plus sub-nodes extending its functionality. AI agent nodes with attached models, memory, and tools are the common example.

A workflow called by another workflow. It runs as its own execution and returns its result to the caller.

A self-hosted configuration where a main instance serves the editor and enqueues work while separate worker instances execute it, coordinated through Redis.

An AI system given a goal and a set of tools that decides for itself which tools to call and in what order. Unlike a chain, it can use persistent memory.

A fixed sequence of calls to a language model and related components. Chains have no persistent memory, so they cannot reference earlier turns of a conversation.

A resource an AI agent can call to fetch information or take action — in n8n, usually another node, such as an HTTP request or a database query.

Storage that lets an AI agent carry context across interactions. In-memory variants reset on restart; database-backed ones persist.

A numerical vector representing a piece of data, letting AI systems compare meaning rather than exact text. Stored in a vector store.

A database built to store and search embeddings, used to retrieve the passages most relevant to a question.

A technique that retrieves relevant documents at question time and passes them to a language model as context, so it can answer about information it was never trained on.

An AI model producing confident output unsupported by its sources. In RAG systems the usual cause is retrieval that returned the wrong passages, not the model itself.

How closely a model's response reflects its source material. Grounded answers cite what was retrieved; ungrounded ones speculate past it.

The licensing model n8n uses. The source is available and self-hosting is free, but the Sustainable Use License restricts reselling n8n as a hosted service to third parties.