Workflows Overview

Overview

Arke's workflow system lets you build automated processing pipelines that scale from a single operation to thousands of concurrent jobs. The system has two layers:

When to Use Each

Use a klados directly when you need a single operation:

• Process one entity through a specific service
• Batch-invoke the same operation across multiple entities
• Test a new processing agent in isolation

Use a rhiza when you need multi-step processing:

• Chain operations together (e.g., OCR → extraction → description)
• Fan out to parallel workers and reconverge results
• Route items to different processing paths based on properties

Prerequisites

Before working with workflows, you'll need:

1. Authentication — An API key or JWT token (Authentication guide)
2. A collection — Workflows operate on entities within collections (Collections guide)
3. Target entities — The entities you want to process

Two-Phase Invocation

Both kladoi and rhizai use a two-phase invocation pattern:

1. Preview (confirm: false) — Returns what permissions will be granted and validates the request, without executing anything
2. Execute (confirm: true) — Creates the job collection, grants temporal permissions, and starts execution

This lets you review what will happen before committing.

Guides