A workflow builder that connects deterministic data flow with model calls, agents, knowledge, and human review.
Workflow & automation · Documented · Documentation checked 2026-10-04
n8n Cloud or a maintained self-hosted instance; models and connected apps are external services.
Instance version, edition, hosting, integration credentials, and model provider access affect capabilities.
What the platform is for
n8n is useful for moving information through an arranged process. It can add AI where language interpretation or tool choice helps. A beginner should first build a tiny workflow without an agent, inspect each node’s records, and understand credentials before choosing dynamic behavior.
Know the product boundary
n8n Cloud and self-hosting have different maintenance responsibilities. Model providers, knowledge stores, and app connectors are separate services with their own credentials and billing. Current AI Agent uses Tools Agent behavior; old agent-type selectors are legacy.
How the components connect
Triggers start work. Nodes transform records. Expressions map data. A model can classify or draft within a fixed route; an agent chooses among exposed tools. Knowledge retrieval supplies evidence. Human review can pause a tool call before action. Evaluations check answer quality while error paths handle execution failures.
A beginner starting point
Begin with a manual trigger, invented data, and one processing step. Add validation and inspect output. Next use a model on a bounded classification task, then a read-only tool agent. Learn human review, errors, and hosting ownership before activating schedules or consequential tools.
Limits and data choices
A technically successful execution can still contain a wrong answer. Retries can duplicate writes. Logs may retain sensitive input. Self-hosting requires updates, backups, access control, and incident handling. This edition covers the core patterns rather than every node, hosting topology, or enterprise integration.
The technical path
Use official n8n node and evaluation documentation for exact operations and current instance support. Developers can extend nodes, APIs, and deployment under a separately maintained design. Prefer a fixed route when inputs, steps, and outputs are known; use an agent only for genuinely variable tool selection.
Feature map
Triggers, nodes, credentials, and data
Build a deterministic workflow and inspect the records passed between its steps.
AI models, agents, tools, and memory
Connect a model to a bounded tool-using agent and distinguish its memory from business records.
Knowledge retrieval and MCP connections
Choose between an approved knowledge source, a fixed integration tool, and a broader MCP server.
Human review before tool actions
Pause an agent tool call and review its exact action and parameters.
Evaluation, errors, and execution history
Test output quality and operational failures as separate parts of reliability.
Hosting, credentials, and governance
Account for the data and operational responsibilities of Cloud or self-hosting.
Complete workflows
A fixed classification-to-review workflow
Transform a fictional workshop request into a validated private review record.
A FAQ tool agent with action approval
Answer approved workshop questions and pause before any sensitive tool action.
Coverage and availability
Models, modes & conversation
Covered: Model provider / agent choices; not a consumer chat subscription.
Search, research & sources
Covered: Knowledge retrieval and tools; citation correctness must be tested.
Files, analysis & code
Not established: Records, expressions, nodes, and execution data; broader code node reference deferred.
Images, video, audio & voice
Covered: Media operations depend on connected providers; no universal native media generator claimed.
Projects, instructions & memory
Covered: Agent memory differs from persistent business records.
Editing surfaces & deliverables
Covered: Workflow canvas and inspected output; generated media / docs depend on services.
Agents, skills, plugins & MCP
Covered: AI Agent, tool nodes, and MCP server registry; credentials required.
Orchestration & scheduling
Covered: Triggers, fixed routing, selective human review, errors, and stopping rules.
APIs, SDKs, deployment & evaluation
Separate product: Evaluation and hosting responsibilities; every API / custom node path not taught.
Sharing, privacy & governance
Covered: Cloud / self-hosting, credentials, run-data retention, and operator ownership.
Plans, devices & accessibility
Not established: Edition / version / hosting affect access; full plan limits and accessibility not established.
Sources and verification
Documentation checked 2026-10-04. Examples are illustrative unless described otherwise.
Build your first workflow · n8n
Integrate AI · n8n
AI Agent node · n8n
MCP server connections · n8n
Human-in-the-loop for tools · n8n
Test and improve AI workflows · n8n
Error Trigger node · n8n
Stop And Error node · n8n
All executions · n8n
Security audit · n8n