Agent Nodes and Tool Use

Reason, call a tool, observe, repeat

An agent (an agent app or an agent node inside a workflow) gives a model a goal, instructions and a set of tools, and lets it decide which tools to call in a loop until it can answer, using strategies such as function calling or ReAct. Agents are flexible for open-ended tasks but less predictable than fixed workflows. Limit the tools to what the task needs, set a maximum number of iterations, require approval for actions that change data, and log every tool call. If the steps are always the same, a fixed workflow is better. Dify's node names, menus and options change between versions; check the current Dify documentation.

Workflow or agent?

Prefer the more predictable option when it works.

situation                                        choose
steps are always the same                        workflow (fixed graph)
few known branches                                workflow + if/else / classifier
open-ended research across several tools         agent with iteration limit
actions that change data                         workflow, or agent + approval step

Set an iteration limit

A maximum number of agent steps prevents runaway loops and surprise costs.

त्वरित जाँच: When is a fixed workflow better than an agent?

• Never
• When the task is completely open-ended
• When you want less predictability
• When the steps are known and always the same