AgentRun Introduces DSL for Integrating AI Agents into Workflows

AgentRun DSL for Workflow Integration

AgentRun has introduced a Domain Specific Language (DSL) designed to incorporate existing AI agents into defined workflows. This DSL allows for the creation of repeatable steps within agent-powered applications, providing a structured approach to agent operations.

Decision Making with Jev

The DSL utilizes Jev for focused decision-making processes within these workflows. Agents can be called upon for investigation when complex tasks arise, while the application maintains its existing tools, model access, permissions, and budgets.

Technical Requirements and Demo

To use AgentRun, Node 22.19+ and npm are required. A quickstart guide, documentation, and examples are available. A ticket-routing demo, which uses scripted decisions and does not require an API key, is provided for initial exploration. The support example demonstrates a workflow with scripted tools and model responses, including escalation paths based on defined criteria and confidence thresholds.

Workflow Definition

Workflows can be written as JSON. The system uses nodes like 'call' for tool execution and 'judge' for decision-making, referencing schemas like 'Candidate' and 'Fit'. The 'Fit' schema defines criteria for decisions, and the system reads decision outcomes and confidence levels to determine subsequent steps.