Skip to main content

Workflows

📖 Best for: All roles (business / operations / developers) — people who want to chain AI calls, HTTP requests, code execution, conditional branches, and loops into automated pipelines on a drag-and-drop canvas

📖 Reading time: 4 minutes

📖 In one sentence: YingCore's visual orchestration canvas — a game-blueprint-style flow editor for connecting AI calls, HTTP requests, code execution, conditional branches, and loops into automated pipelines, supporting manual, scheduled, and Webhook triggers with full run history and real-time monitoring. Accessible from the top nav "工作流 / Workflows" (route /dynamic-workflows) for all roles. For advanced usage see Advanced Workflow Orchestration.

1. Capabilities​

1. Visual canvas + DSL dual view​

Drag-and-drop nodes and wires to build flows. Supports two edge types: execution edges (control flow order) and data edges (pass parameters between nodes). Switch between canvas and DSL code view with one click. Import / export JSON, undo / redo, simple / advanced mode all included.

2. 30+ node types​

CategoryNodes
Flow controlStart, End, Trigger, Switch branch, If gate, Categorize, Parallel, Loop, Iteration, Invoke sub-workflow, Master Agent, Read / Write global variables
AILLM call (configure prompt / model / temperature)
NetworkHTTP request
Compute / DataCode execution, data transform, list / dict operations, string handling, message output
OperatorsAND / OR / NOT, comparisons (>, <, =), add / subtract / multiply / divide
Type conversionConvert to string / number / boolean / array / object
TimeCurrent time, format / parse time
ConstantsString, number, boolean, array, object, date
OtherDelay, debug checkpoint, Skill nodes (dynamically loaded from Skill Marketplace), graph analysis, graph query

3. Four trigger modes​

ModeDescription
ManualClick 「Run」 on the list / detail page to execute immediately
Scheduled (cron)Periodic execution on cron expression; one-click enable / disable on the list page
One-shot scheduledExecute once at a specified time, then auto-revert to manual mode
WebhookTriggered by HTTP request; auth header supported

4. Variables and context​

  • Pass parameters between nodes via ${variable} syntax. Type ${ to trigger the variable-aware input popup.
  • The top 「Global Variables Panel」 centralizes shared variables across nodes.
  • Supports JSON path extraction (e.g. ${user.profile.name}) and function transforms.

5. Debugging and run monitoring​

  • Test run: starts in test mode without polluting real data.
  • Breakpoint debugging: pause at a node to inspect context.
  • SSE real-time monitor: node status pushed to the monitor page in real time.
  • Error details: expand an errored node to see stack trace, inputs / outputs, and context variables.
  • Re-run failed node: click 「Re-execute」 on a failed node in run history to retry only that node.

6. Run history and comparison​

  • Stats cards: total runs / success rate / success & failure counts / average duration.
  • Status filter: success / failure / running / cancelled.
  • Vertical timeline: one card per run.
  • Diff comparison: select two runs to diff—issues stand out instantly.

7. Version control​

Every save generates a version number. Roll back to any historical version with one click. Combined with import / export JSON, you can share workflow templates across teams.

2. Typical Scenarios​

Scenario 1: Auto-assign customer support tickets​

Webhook receives a ticket → LLM node identifies intent → Switch routes by ticket type → dispatches to the matching digital employee → on failure, hands off to a human agent. Response time cut from 10 minutes to 30 seconds.

Scenario 2: Daily sales data sync​

Cron triggers at 8:00 AM → HTTP fetches ERP sales data → data transform cleans it → writes to knowledge base → pushes daily report to Feishu group. Zero human touch; data lands by 8:05 AM the next day.

Scenario 3: Marketing campaign launch​

After a plan is approved in Marketing Plan Center, a workflow is auto-created → end-to-end execution of acquisition, outreach, and data collection. Plan to execution with zero wait.

Scenario 4: Multi-model comparison & evaluation​

The same prompt runs in parallel across multiple LLMs via the Parallel node → data transform unifies formats → LLM scoring node picks the best → writes to database. Typically used for pre-release prompt A/B evaluation.

Scenario 5: Marketing digital employee collaboration​

Use the 「Custom Marketing Workflow」 at the /templates entry: AI creator generates copy → human review → AI push → sales follow-up. The whole flow is orchestrated visually on the canvas.

3. UI & Operations​

1. Workflows list /dynamic-workflows​

AreaContent
TopSearch box + 「New Workflow」 button
CardsOne card per workflow, menu: Edit canvas / View details / Run history / Delete
Quick actionsRun now, toggle scheduled switch, edit description directly on the card

2. Workflow detail /dynamic-workflows/detail/:id​

  • Basic info: name, description, current version, last modified time.
  • Canvas stats: node count / wire count / skill count.
  • Action bar: 「Run Now」, 「Edit Canvas」, 「Delete」.
  • Recent 5 runs: click to enter the real-time monitor page.

3. Canvas editor /dynamic-workflows/new or /:id​

The core orchestration interface with a 5-zone layout:

ZoneFunction
Left node panelNode categories collapsed, drag onto the canvas
Center canvas / DSL viewDrag-and-drop orchestration; switch to DSL code view with one click
Right property panelConfigure the selected node's parameters; variable-aware input supported
Top toolbarSave, validate, run, test, import / export JSON, copy, undo / redo, simple / advanced mode
Bottom debug consoleReal-time logs and node outputs

Advanced features: variable-aware input, global variables panel, test run dialog, breakpoint debugging.

4. Run history /dynamic-workflows/:workflowId/runs​

  • Top stats cards: total runs / success rate / success & failure counts / average duration.
  • Status filter: success / failure / running / cancelled.
  • Vertical timeline: one card per run (start time, duration, node count).
  • Diff comparison: select two runs to diff.

5. Real-time monitor /dynamic-workflows/runs/:runId​

  • Top status bar + progress bar.
  • Node execution card list: per-node status and details (collapsible); SSE-pushed live updates.
  • Click any node to expand its inputs, outputs, context, and error stack.

6. Marketing workflow templates /templates​

The Workflows module embeds a set of marketing-domain workflows (independent API, parallel with v2 dynamic workflows):

  • AI digital employee card wall: showcases content generation, content push, info collection, intelligent customer service, and the human (reviewer / salesperson) collaboration steps.
  • Custom marketing workflow: nodes are pre-set with roles like AI Creator / AI Push / AI Customer Service / Human Reviewer / Sales. Supports cron scheduling, copy selection, and personnel assignment. Once published, runs automatically per the flow.

See Digital Employees and Marketing Plan Center for more on marketing orchestration.

4. How to Use​

Create and orchestrate​

  1. Go to 「Workflows」 → click 「New Workflow」.
  2. Pick a starter template (e.g. *「Scheduled bulk skill dispatch」**) or 「Blank canvas」; fill in name and description.
  3. Drag nodes from the left panel onto the canvas.
  4. Connect nodes with execution edges for flow order; add data edges wherever you need to pass parameters.
  5. Configure node parameters in the right property panel; type ${ to trigger variable-aware input.
  6. Click 「Save」 on the toolbar (auto-save supported); click 「Validate」 to check configuration correctness.

Tip: Name nodes semantically (e.g. 「Identify ticket intent」)—you can locate problem nodes at a glance in run history. Always click 「Validate」 before running externally imported JSON.

Run and monitor​

  1. Click 「Run」 on the canvas toolbar, or click 「Run Now」 on the list / detail page.
  2. The real-time monitor page opens automatically—watch each node's progress.
  3. After completion, go to the run history page for stats and timeline. Select two runs to diff; expand any errored node for details.

Scheduled workflows​

  1. Set a cron expression in the workflow's trigger configuration.
  2. After saving, toggle the workflow's enable switch on the list page.
  3. The system runs automatically on the cron schedule. You can disable at any time.

Debug and troubleshoot​

  1. Click 「Set breakpoint」 on a suspicious node.
  2. Click 「Test」 on the toolbar to run in test mode.
  3. Inspect inputs, outputs, and context variables at the paused breakpoint.
  4. Expand errored nodes for stack details; click 「Re-execute」 on a failed node in run history to retry only that node.

Version rollback​

  1. Go to the workflow detail page → 「Version History」.
  2. Select any historical version → 「Rollback to this version」.
  3. The system creates a new version (keeping history) and applies it immediately.

5. Tips​

  • Name nodes semantically: 「Identify ticket intent」 beats 「LLM 1」—locate problem nodes in run history at a glance.
  • Prefer data edges over hard-coding: Pass parameters via data edges whenever possible to make upstream data easier to swap later.
  • Use global variables sparingly: Changing a global variable affects every node referencing it. When debugging, temporarily comment out the suspicious reference.
  • Test run before production run: For complex flows, run 「Test」 to debug first, then 「Run」 for real data.
  • Re-run failed nodes individually: Click 「Re-execute」 on a failed node in run history to retry only that node, without affecting others.
  • Always auth Webhooks: When enabling Webhooks in production, configure an X-Token auth header to prevent arbitrary triggering.
  • Mind cron timezones: Scheduled workflows run on the server's timezone by default. For cross-region teams, account for timezones explicitly in the cron expression.
  • Use the dedicated entry for marketing flows: When orchestrating marketing scenarios, prefer the 「Custom Marketing Workflow」 at /templates. Nodes are pre-set with AI Creator / Reviewer roles—3-5x faster than the generic canvas.