Workflow Engine

Yunque Agent includes a visual workflow engine for automating multi-step processes. Workflows define a DAG (Directed Acyclic Graph) of nodes connected by edges, executed by the runtime engine.

Concepts

• Definition: a reusable workflow template with nodes and edges
• Instance: a running execution of a definition
• Node: a single step (LLM call, code execution, browser action, etc.)
• Edge: a connection between nodes defining execution order

Node Types

Type	Description	Executor
start	Entry point	Built-in
end	Exit point	Built-in
llm	LLM completion with configurable model/prompt	Planner
code	Execute Python/JS/Go/Shell code	Sandbox Runner
browser	Browser automation (navigate, click, extract)	BrowserEngine
knowledge	Knowledge base hybrid search	KnowledgeStore
condition	Conditional branching (if/else)	Built-in
loop	Repeat a subgraph N times	Built-in
api	External HTTP API call	HTTP client
transform	Data transformation (JSON/text)	Built-in
wait	Pause for duration or event	Built-in
human	Wait for human input	Built-in
subworkflow	Invoke another workflow	Recursive
parallel	Execute branches in parallel	Built-in

Creating a Workflow

Via API

POST /v1/workflows
Content-Type: application/json

{
  "name": "Daily Report",
  "description": "Generate a daily summary report",
  "nodes": [
    { "id": "start", "type": "start" },
    { "id": "fetch", "type": "api", "config": { "url": "https://api.example.com/metrics" } },
    { "id": "analyze", "type": "llm", "config": { "prompt": "Summarize these metrics: {{fetch.output}}" } },
    { "id": "end", "type": "end" }
  ],
  "edges": [
    { "from": "start", "to": "fetch" },
    { "from": "fetch", "to": "analyze" },
    { "from": "analyze", "to": "end" }
  ]
}

Via Dashboard

1. Navigate to Workflows in the sidebar
2. Click New Workflow
3. Use the visual node editor to drag-and-drop nodes
4. Connect nodes with edges
5. Configure each node's parameters
6. Save and run

Running Workflows

POST /v1/workflows/:id/run

{
  "inputs": {
    "date": "2026-04-12"
  }
}

Monitoring Instances

GET /v1/workflows/:id/instances     # List instances
GET /v1/workflows/:id/instances/:iid  # Instance detail
POST /v1/workflows/:id/instances/:iid/cancel  # Cancel

Storage

Workflow definitions and instances are persisted to Ledger KV with file-system fallback. Data survives restarts.

API Reference

Method	Path	Description
GET	/v1/workflows	List all definitions
POST	/v1/workflows	Create definition
GET	/v1/workflows/:id	Get definition
PUT	/v1/workflows/:id	Update definition
DELETE	/v1/workflows/:id	Delete definition
POST	/v1/workflows/:id/run	Start instance
GET	/v1/workflows/:id/instances	List instances
POST	/v1/workflows/:id/instances/:iid/cancel	Cancel instance