Agent · Workflow · Chatflow

Evose offers three building primitives. Their capabilities overlap, but choosing the wrong one will make your build over-complex or hard to maintain.

One-line Distinction

Primitive	Form	Mental model
Agent	Conversation	"I talk to a role and it responds."
Workflow	Process	"I provide input, the system runs predefined steps to produce output."
Chatflow	Conversation + orchestration	"I talk to a conversation flow that has node-based orchestration."

Chatflow is the "advanced form" of Agent — when a conversation needs multi-node decisions, upgrade a prompt-only Agent to a Chatflow Agent.

Decision Tree

Does your scenario need multi-turn conversation?
├─ Yes → Does it need complex branching / tool calls / data orchestration?
│       ├─ No  → Prompt-based Agent
│       └─ Yes → Chatflow Agent
└─ No → Is it a fixed-flow batch / scheduled / automation task?
        ├─ Yes → Workflow
        └─ No  → Reconsider your needs — you may not need Evose

Detailed Comparison

Dimension	Prompt Agent	Chatflow Agent	Workflow
Form	Single prompt + tools	Visual node canvas + chat	Visual node canvas + I/O
Trigger	User chat	User chat	Manual / scheduled / API / webhook / conditional
State	Single- or multi-turn context	Multi-turn + node state	Single execution or scheduled
Nodes	None (only LLM + tools)	9 groups, 19 nodes (incl. interaction)	8 groups, 17 nodes (no interaction)
Typical scenarios	Customer service / assistant / translation	Multi-turn form intake / complex CS / guided dialog	Article generation / data ETL / report automation
Learning curve	5 minutes	30 minutes	30 minutes

Chatflow's 9 Node Groups

Group	Typical nodes	Purpose
AI	LLM · Knowledge retrieval	Call models, do RAG
Logic	Condition · Batch · Loop · Intent	Branching and looping
Data	Code · Template · Variable · Document extraction · Parameter extraction	Process structured data
Network	HTTP	Call external APIs
App	Agent · Workflow	Call other apps
Tool	MCP · Plugin · Marketplace tool	Plug in external capabilities
Interaction (Chatflow only)	Direct reply · Form input	Interact with the user
End	End	Terminate

Workflow has the same node structure but no Interaction group (it doesn't talk to a user in real time).

"Apps Calling Apps" Is Allowed

A Workflow can call an Agent for a single reasoning step
A Chatflow can trigger a Workflow as a background task
A Workflow can orchestrate multiple Agents in series

This lets you layer abstractions: each base primitive owns one responsibility, while the upper-layer flow does orchestration.

When to Upgrade or Split

Symptom	Signal	Suggestion
Prompt grows too long (> 500 chars)	Complexity overflow	Upgrade to Chatflow, split decisions into nodes
One Workflow has > 20 nodes	Hard to maintain	Split into sub-Workflows, orchestrate from a parent
The same prompt is reused in many places	Duplication	Extract into a Skill
The same external API is called in many places	Duplication	Extract into an organization-level Tool

Don't Do This

Agent doing heavy data processing — stuffing "process a giant CSV" into an Agent. That's a Workflow scenario.
Workflow doing dialog — back-and-forth user interaction shouldn't be forced into a fixed flow.
Chatflow for trivial prompts — don't draw a 10-node diagram for something a single prompt solves.
Cross-app loops nested deeply — Workflow → Agent → Workflow → Agent will make Traces unreadable and costs hard to attribute.

Next Steps

Picked Agent → Create an Agent
Picked Workflow → Create a Workflow
Want real-world scenarios → Customer service bot · Marketing automation

Agent · Workflow · Chatflow

On this page