Workflows are visual automations that define exactly how Duckie handles specific scenarios with deterministic, step-by-step execution.Documentation Index
Fetch the complete documentation index at: https://docs.duckie.ai/llms.txt
Use this file to discover all available pages before exploring further.
What is a Workflow?
A workflow is a visual graph of connected nodes that define a precise execution path. Each node performs an action or makes a decision, and edges connect nodes to define the flow. Unlike runbooks (which give agents flexibility), workflows follow explicit branches and conditions every time.When to Use Workflows
Workflows are ideal when:| Situation | Why Workflows Work |
|---|---|
| Exact steps required | Every execution follows the same path |
| Compliance/audit needs | Deterministic execution is traceable |
| Complex branching | Multiple decision points with clear outcomes |
| Tool orchestration | Multiple tools must coordinate precisely |
Workflow Components
Nodes
Nodes are the building blocks of workflows:| Node Type | Purpose | Example |
|---|---|---|
| Start | Entry point for the workflow | ”Begin when the deployment trigger fires” |
| Tool | Execute a Duckie tool, app tool, custom tool, MCP tool, runbook, or workflow | ”Search company docs” |
| Decision | Route to different branches using rule or AI conditions | ”Is the order eligible?” |
| End | Mark a workflow path complete | ”Refund handled” |
Edges
Edges connect nodes and define the flow:- Success path — Where to go when the node succeeds
- Failure path — Where to go when the node fails
- Branch paths — Decision outputs evaluated in order, with an else path when nothing matches
Inputs and Outputs
Each node can:- Accept inputs — Manual values, ticket data, saved values, previous node outputs, or AI-generated values
- Produce outputs — Data passed to subsequent nodes
Example Workflow
Refund Request Workflow: What this does:- Extracts the order number from the customer’s message
- Looks up the order in your system
- Checks if it’s within the return window
- Either processes the refund OR explains the policy
- Sends the appropriate response
Workflows vs Runbooks
| Aspect | Workflows | Runbooks |
|---|---|---|
| Format | Visual graph | Natural language |
| Execution | Deterministic — same path every time | Flexible — AI adapts |
| Best for | Strict processes, compliance | Judgment-based conversations |
| Complexity | Visual but more setup | Easy to write |
| Debugging | See exact path taken | See AI reasoning |
Choose Workflows When
- Process must be identical every time
- Audit trails require deterministic paths
- Multiple tools must execute in exact order
- Branching logic is complex but well-defined
Choose Runbooks When
- Conversations vary significantly
- AI judgment improves outcomes
- Quick iteration is important
- Instructions are easier than flowcharts
Workflow Versions
Workflow edits are saved as versions:| Version type | Meaning |
|---|---|
| Draft | Saved changes that are not live for agents yet |
| Published | The version available for agents to use |
Next Steps
Create a Workflow
Build your first workflow
Nodes & Conditions
Reference for all node types
Compare to Runbooks
Understand when to use each