Workflows

Workflows orchestrate multi-step operations across agents, tools, and knowledge assets to deliver repeatable, governed outcomes

Workflows standardize how complex tasks are performed—from data collection and enrichment to analysis, drafting, review, and publishing—while enforcing policies and capturing provenance. They encode proven processes so outcomes are consistent and auditable.

Core Concepts

• Workflow: A named, versioned process definition describing steps, transitions, and policies.

• Step: A unit of work (human or automated) that transforms inputs into outputs.

• Context: The execution state containing conversation/data-room references, documents, insights, and parameters.

Workflow Execution Mode

There are 2 modes for workflows: run_all_step and step_by_step, corresponding to Autopilot and Human in the Loop mode respectively.

For API Access, it is recommened to use run_all_step mode for easier integration.

Best Practices

• Start with a minimal viable path; add branches and gates incrementally.

• Encapsulate reusable steps and evaluation criteria for consistency.

• Prefer parallelism for independent tasks; guard with resource limits.