Connect a model to a bounded tool-using agent and distinguish its memory from business records.
Workflow & automation · Documented · Documentation checked 2026-10-04
n8n Cloud or a maintained self-hosted instance; models and connected apps are external services.
Provider credentials and compatible instance version required. Current AI Agent uses Tools Agent behavior; older agent-type selectors are legacy.
The worked example is illustrative and has not been tested in a signed-in product account.
What it does and how it connects
Connect a model to a bounded tool-using agent and distinguish its memory from business records.
An AI Agent node combines a chat model with tools and optional memory. Current documentation uses Tools Agent behavior and requires at least one connected tool. Older agent-type menus are legacy. Memory supports conversational context; it is not a trusted database of business facts.
Walkthrough
Use the product named in the access panel. The steps below combine documented controls with the editorial example.
- Build a manual sample workflow before adding AI.
- Connect an eligible chat model with provider credentials and at least one narrowly scoped tool.
- Write a goal that names allowed sources, output, and stopping conditions.
- Inspect agent activity and tool inputs, then validate the final output outside the prompt.
Worked example
Give a sample agent one read-only tool for an approved workshop FAQ. It should answer known questions and route absent answers to a person. Avoid tools that send invitations while testing retrieval quality.
Answer from the approved workshop FAQ tool only. Cite the supporting entry. If no entry answers the question, return needs_review with the missing detail. Never invent dates, fees, or registration status. Do not use unrelated tools or take external write actions.
Inspect and verify the result
Inspect the tool chosen, parameters, returned evidence, and final answer. Test a known question, an absent answer, and an instruction attempting to broaden the task. Validate output fields in a separate workflow step.
Limits and troubleshooting
Models can choose the wrong tool or repeat calls. An agent needs a tool connection even if your prompt sounds self-contained. Bound iterations and resources using supported controls; report provider or tool failures separately from answer uncertainty.
Data, permissions, and actions
Model calls send selected context to the provider, while tools act with their own credentials. Limit operations and source scope. A system prompt cannot grant safe isolation or prove that action permissions are narrow.
Related reading
Knowledge retrieval and MCP connections
Choose between an approved knowledge source, a fixed integration tool, and a broader MCP server.
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.
Permissions and meaningful approvals
Decide what a system may read, what it may change, and when someone needs to authorize a concrete action.
Sources and verification
Documentation checked 2026-10-04. Examples are illustrative unless described otherwise.
Integrate AI · n8n
AI Agent node · n8n