n8n & automation glossary.

An AI agent in n8n is a node (typically @n8n/n8n-nodes-langchain.agent) that wraps an LLM in a loop with tool-calling. It plans, executes a tool, observes the result, and repeats until done. Used for tasks where the model needs to make decisions across multiple steps. See agent-loop workflows in the catalog.

Application Programming Interface — the contract that lets one piece of software call another. Most n8n integrations talk to remote APIs over HTTPS via either a dedicated node or the generic HTTP node.

A point where the graph splits into two or more paths based on a condition. In n8n the IF or Switch node creates branches; the Merge node rejoins them.

A scheduled trigger that fires on a time pattern (every 5 minutes, every Monday 09:00, etc). The Cron node is one of n8n's two main trigger types — see also Webhook.

Stored authentication data for an integration — API key, OAuth token, basic auth, etc. n8n stores credentials encrypted; AutomationFlows strips credential IDs before publishing — see the full methodology.

A community- or user-built integration that extends n8n. Distributed via npm packages with the n8n-nodes- prefix.

A numeric vector representing the meaning of a chunk of text. Generated by an embedding model (OpenAI ada, Cohere, local) and stored in a vector DB for similarity search. Backbone of RAG pipelines.

A single run of a workflow. n8n stores execution history with the inputs, outputs, and any errors per node.

An n8n node that runs arbitrary JavaScript on each item flowing through. Used when the built-in expression syntax isn't enough — the universal code escape hatch.

n8n's universal API escape hatch — call any HTTP endpoint that doesn't have a dedicated integration node. Supports auth, body, headers, retries. Browse HTTP-node workflows.

Branches the workflow into a true/false path based on a condition expression. The simplest form of branching.

n8n stores every workflow as a JSON file describing nodes (with type, parameters, position) and connections. The JSON is portable — paste into another n8n and it runs. The full import-by-JSON guide covers the workflow.

A framework for chaining LLM calls with tools, memory, and retrieval. n8n's LangChain integration nodes wrap most of LangChain's primitives. See AI & RAG workflows.

A workflow that runs only when a user clicks 'Execute workflow'. Useful for ad-hoc tools and admin scripts.

Open-source workflow automation platform — visual editor for connecting integrations with logic and AI. Self-hostable, JSON-portable. Pronounced 'n-eight-n' (eight letters between the two n's). Compare against the alternatives: vs Zapier, vs Make, vs Activepieces.

The hosted version of n8n run by n8n GmbH. Removes the self-hosting overhead at the cost of monthly per-execution pricing.

A single step in an n8n workflow. Each node has a type (which integration / logic primitive) and parameters. Workflows are graphs of nodes.

An n8n node for calling OpenAI's chat / embedding / image APIs. Often paired with the LangChain Agent node for tool-using AI workflows. Browse OpenAI workflows.

Each node receives an input array of items and produces an output array of items. The arrays flow downstream through the connections.

Test-run data n8n records on each node so you can re-run with the same input. AutomationFlows strips pinData before publishing because it can contain real API responses — see the methodology for the full privacy strip.

A trigger that periodically checks an external system for new data — RSS feed updates, new Airtable rows, new Stripe customers. Different from webhook triggers in that the n8n side initiates each check.

An LLM pattern: retrieve relevant docs from a vector DB, inject them into the prompt, then ask the LLM. Reduces hallucination and lets the model answer from your data. Browse RAG ingestion workflows.

Running n8n on infrastructure you own — Docker on a VPS, Kubernetes, etc. Trade-off: ops work in exchange for fixed cost and full data ownership. The main reason to choose n8n over Zapier.

An n8n graph element that's just text — used to document a workflow visually. AutomationFlows scores sticky-note density as part of the QualityScore.

A workflow that's invoked as a node inside another workflow. Lets you factor out reusable logic the same way functions do in code.

Like IF but with N output branches. Used when routing on enum-like values (priority levels, region codes, status names). See IF node for the simpler 2-way version.

The starting node of a workflow. Defines when it runs: webhook, cron, manual click, app event (new Stripe customer, new Slack message), polling.

A database optimised for similarity search on embeddings. Pinecone, Qdrant, Weaviate, Supabase Vector, pgvector. Backbone of RAG pipelines. Browse Pinecone / Qdrant / Supabase Vector workflows.

An HTTP endpoint that an external system calls when something happens. n8n's Webhook trigger gives you a unique URL to drop into Stripe / GitHub / form services. Browse webhook-handler workflows.

A directed graph of nodes connected by data flow. The unit of automation in n8n. Each catalog entry on AutomationFlows is one workflow.

A single record flowing between nodes. Each node receives an array of items and produces an array of items. The shape is {json: {...}, binary: {...}}. See also Output / Input.

Inline JavaScript in any field, wrapped in {{ ... }} or referenced via ={{ $json.field }}. Used to pull data from previous nodes, transform values, or compute conditionals without writing a Function node.

Joins two branches back into one. Modes: append (concatenate), pass-through, multiplex (cartesian product), combine-by-key. Pair with the IF or Switch node when a workflow needs to fork then rejoin.

An n8n node that creates, modifies, or filters fields on the items flowing through. The mainstay of inline data transformation — used in roughly half of all workflows in the catalog.

Runs JavaScript or Python on each item or on all items at once. The 'code escape hatch' when expressions or built-in nodes can't express the logic. Distinct from the legacy Function node — the Code node is the modern replacement.

Processes a large input array in batches of N items, then loops back. Used to rate-limit API calls or process huge result sets without holding everything in memory at once.

Inverse of SplitInBatches: takes one item with an array field and emits one output item per element of that array. Common pattern when an upstream API returns a list and downstream nodes need to act on each element.

A modern n8n node that iterates over each item in the input. Replaces the older SplitInBatches pattern for many use cases — clearer semantics, explicit done condition.

The core node for tool-using AI in n8n. Wraps an LLM with one or more tool nodes connected as inputs; the agent picks tools, calls them, observes results, and iterates. Browse agent-loop workflows.

A node that an AI Agent can invoke during its loop. Examples: HTTP Request Tool, Calculator Tool, Workflow Tool (which calls a sub-workflow), Custom Tool. Tools have descriptions the LLM reads to decide when to use them.

Stores conversation context between turns when chatting with an AI Agent. Variants: BufferWindow (last N turns), Postgres-backed, Redis-backed. Without memory, every turn is independent.

Same as Vector DB but with LangChain-specific n8n nodes that handle ingestion, retrieval, and chain-integration uniformly across providers (Pinecone, Qdrant, Weaviate, Supabase pgvector, in-memory).

Forces an LLM response into a known structure (JSON schema, regex, Pydantic-style). Without one, you're parsing free-form text. Available output parsers in n8n: Structured (JSON schema), Auto-fixing, Item List.

Loads raw documents (PDF, JSON, web page, file binary) into the format the chunker / embedder expects. First node in a typical RAG ingestion pipeline. Browse vector-store-ingest workflows.

Chunks long documents into LLM-context-sized pieces. Strategies: recursive character (split on paragraph → sentence → word), token-aware, character-fixed. Critical for RAG quality.

A built-in n8n form that produces a public URL collecting structured input. Each submission triggers the workflow. Often paired with downstream LLM nodes for "AI form" patterns. See Form Trigger workflows.

Spawns a chat-style web interface attached to a workflow. Each user message starts an execution; the workflow's final output is shown back to the user. The standard front-end for n8n-hosted AI assistants.

A separate workflow designated to run whenever ANY workflow in the n8n instance fails. Set per-workflow under Settings → Error workflow. Common pattern: Slack alert + log to Postgres + retry queue. See error-handling tutorial.

Property of an operation that produces the same result whether run once or many times. Important in n8n because retries are common — a non-idempotent webhook handler could create duplicate records on a retry. Achieve via dedup keys, conditional creates, or upserts.

Per-instance secret n8n uses to encrypt credentials at rest in its database. If lost, every credential becomes unreadable and must be re-entered. Backup target #1 when self-hosting. See self-hosting tutorial.

A scaling architecture where one main n8n process handles incoming events and dispatches workflow executions to N worker processes via Redis. Required when you need parallelism beyond what a single Node.js process can provide. See deploy-on-hetzner tutorial for the queue-mode setup.