Workflows and agents - Flowra

This is the canonical developer explanation of workflow vs agent. Product-level summary: How Flowra works. Native tool slugs: Native tools: Workflow. API CRUD and execute: Workflows.

Same record, two behaviors

In Flowra both are stored as a workflow with workflowType:

	Workflow (`static`)	Agent (`agent`)
Behavior	Fixed graph of steps (nodes and edges): actions, code, optional `llm_agent` nodes	Conversational AI: system prompt + tools; the model chooses tools each turn
Best for	Scheduled jobs, ETL, “always A → B → C”, approvals (`human_wait`)	Chat UIs, support bots, open-ended tasks
How you run it	`POST /workflow/manager/execute/{id}`, cron, or trigger tools	Dashboard chat, embeds, or Threads / Graphify stream — not the static execute path for chat turns
Triggers	`manual`, `cron`, or `trigger_tool`	`manual` (chat) or `trigger_tool` for inbound events

A static workflow can still contain llm_agent nodes — AI runs only inside that step, while the overall path is your graph. An agent has no fixed top-level graph.

Which path should you use?

Need a human to approve a step? See Human in the loop.
Building a chat UI? See Threads.
Exposing tools to Cursor? See MCP.
Calling from your backend? See SDKs (workflows.* for static execute, chat.* for agent streams).

Workflows — List, create, execute, executions
Triggers
Skills — Attach skill packs to agents
Native tools: Workflow