AgentRun, an open-source DSL by grep.ai, converts repetitive agent tasks into workflows combining tool calls, code, and LLM judgments to reduce costs and improve clarity.
- •It automatically identifies workflow-convertible parts by analyzing traces and reflection notes left by agents during work.
- •It integrates tool calls, code, LLM routing, evidence screening, and agent execution for additional research into a single DSL.
- •It aims to reduce the cost of full agent process execution while improving workflow readability and maintainability.