Skip to main content

Execution

The execution layer orchestrates LLM calls, data transformations, and validated outputs based on your workflow configuration.

Core Configuration

SettingPurposeOptions
run_modeProcessing strategyonline (real-time), batch (cost-optimized)
granularityProcessing scoperecord (per-item), file (all items)
guardConditional executionSkip or filter based on data
retryError handlingAutomatic retry for transient failures
versionsParallel variationsMultiple iterations with different parameters

Quick Reference

NeedConfiguration
Lower costs, can wait 24hrun_mode: batch
Immediate responsesrun_mode: online
Per-item transformationsgranularity: record
Aggregation/exportsgranularity: file
Conditional executionguard: { condition: "...", on_false: filter }
Handle transient failuresretry: { max_attempts: 3, on_exhausted: return_last }

Schema Analysis

Analyze workflow schemas and field dependencies before making API calls:

agac schema -a my_workflow

Shows input/output schemas for each action and helps catch field reference errors.

See Also