Why Verdog?

Verdog is an agentic workflow programming language built on top of Python. It generates code scaffolding from visually editable workflow graphs, while users define node implementations, agent configurations, and prompts.

Visual workflows, Python implementations

Problem

As agents, checks, and repair loops accumulate, control flow becomes difficult to follow.

Verdog

Inspect the workflow structure in a visual graph and implement each step in Python. Graph edits preserve your implementations, agent configurations, and prompts.

Less orchestration boilerplate

Problem

Workflows repeatedly use common control-flow patterns: branching, iteration, and calls to reusable components. Implementing this orchestration by hand adds repetitive code and maintenance effort.

Verdog

Declare the workflow structure as a graph. Code generators provide the scaffolding, and a shared runtime executes it, leaving users to implement the task-specific behavior.

Early error detection

Problem

Errors in inputs, outputs, and connections between steps may remain unnoticed until execution, wasting time and resources.

Verdog

Generated Python type declarations allow type checking to detect incompatible interfaces before a workflow runs, with diagnostics shown directly in VS Code.

Structural termination analysis

Problem

Agentic loops may repeatedly attempt to improve a result without making progress, consuming time and resources.

Verdog

Certify termination for every input covered by the declared model, provided individual steps terminate and respect their declared conditions and effects.

Termination does not imply that the desired result is achieved.

Reusable components

Problem

Reimplementing the same process across projects duplicates work. Combining existing components can introduce conflicting dependencies.

Verdog

Compose subroutines or reuse workflows with isolated Python environments. Shared workflows are pinned to Git commits for reviewable updates.

Local execution and recovery

Problem

Long-running workflows are costly to restart after interruption.

Verdog

Run workflows locally, inspect their logs and history, and resume or fork from restorable checkpoints. The user owns the source code.

Read Why agentic workflows? for the motivation and limits of verification and corrective feedback.