Workflow¶

Configure workflow orchestration in Datus Agent to define execution plans, node sequences, parallel processing, and sub-workflow composition. Workflows control how nodes are executed to process user queries and generate SQL.

Quick Start

New to workflows? Start with the Basic Workflow Configuration section to understand the fundamentals.

Workflow Structure¶

Workflows are defined in the configuration file under the workflow section:

agent.yml

workflow:
  plan: planA                        # Active execution plan

# Define execution plans
planA:
  - schema_linking
  - generate_sql
  - output

planB:
  - schema_linking
  - generate_sql
  - execute_sql
  - reflect
  - output

Basic Workflow Configuration¶

Simple Sequential Workflow¶

Simple Sequential Workflow

workflow:
  plan: basic_sql

basic_sql:
  - schema_linking                   # Find relevant tables
  - generate_sql                     # Create SQL query
  - output                          # Format results

Workflow with Execution and Reflection¶

Workflow with Execution and Reflection

workflow:
  plan: with_execution

with_execution:
  - schema_linking                   # Find relevant tables
  - generate_sql                     # Create SQL query
  - execute_sql                     # Run the query
  - reflect                         # Analyze results
  - output                          # Format final output

Advanced Workflow Features¶

Advanced Features

The following features require careful planning and understanding of node dependencies. Test thoroughly in development environments.

Parallel Execution¶

Execute multiple nodes simultaneously and then select the best result:

Parallel Execution Example

workflow:
  plan: parallel_generation

parallel_generation:
  - schema_linking
  - parallel:                       # Execute in parallel
      - generate_sql
      - reasoning
  - selection                       # Choose best result
  - execute_sql
  - output

Sub-workflows¶

Break complex workflows into reusable sub-components:

Sub-workflows Example

workflow:
  plan: multi_approach

multi_approach:
  - schema_linking
  - parallel:
      - subworkflow1
      - subworkflow2
      - subworkflow3
  - selection
  - execute_sql
  - output

# Define sub-workflows
subworkflow1:
  - search_metrics
  - generate_sql

subworkflow2:
  - search_metrics
  - reasoning

subworkflow3:
  - reasoning
  - reflect

Sub-workflows with Custom Configuration¶

Each sub-workflow can use its own configuration file:

Sub-workflows with Custom Configuration

workflow:
  plan: multi_agent

multi_agent:
  - schema_linking
  - parallel:
      - subworkflow1
      - subworkflow2
      - subworkflow3
  - selection
  - execute_sql
  - output

subworkflow1:
  steps:
    - search_metrics
    - generate_sql
  config: multi/agent1.yaml          # Custom config file

subworkflow2:
  steps:
    - search_metrics
    - reasoning
  config: multi/agent2.yaml

subworkflow3:
  steps:
    - reasoning
    - reflect
  config: multi/agent3.yaml

Built-in Workflow Plans¶

Datus Agent provides three built-in workflow plans if you don't configure custom ones:

Default Workflows

These workflows are automatically available and can be referenced without additional configuration.

Reflection WorkflowFixed WorkflowMetric-to-SQL Workflow

# Built-in: reflection
reflection:
  - schema_linking
  - generate_sql
  - execute_sql
  - reflect
  - output

# Built-in: fixed
fixed:
  - schema_linking
  - generate_sql
  - execute_sql
  - output

# Built-in: metric_to_sql
metric_to_sql:
  - schema_linking
  - search_metrics
  - date_parser
  - generate_sql
  - execute_sql
  - reflect
  - output

Reflection Nodes Configuration¶

Configure alternative workflows that can be triggered during reflection:

Reflection Nodes Configuration

reflection_nodes:
  # When schema linking needs improvement
  schema_linking:
    - schema_linking
    - generate_sql
    - execute_sql
    - reflect

  # When document search is needed
  doc_search:
    - doc_search
    - generate_sql
    - execute_sql
    - reflect

  # Simple regeneration
  simple_regenerate:
    - execute_sql
    - reflect

  # Enhanced reasoning
  reasoning:
    - reasoning
    - execute_sql
    - reflect