Back to docsRecipe

ReAct Agent Loop

The Reasoning + Acting paradigm that powers Meridian agents. Think, act, observe — repeat until the task is done.

The Loop

  1. 1Thought — the agent reasons about the current state and decides what to do next.
  2. 2Action — it invokes a tool: search, code execution, API call, or file read.
  3. 3Observation — the result is fed back into context, grounding the next thought.

Why it matters

Without a structured loop, agents hallucinate or drift. ReAct anchors every step in observable reality. Meridian agents use this pattern to stay on-rails across multi-turn tool use, long-running research tasks, and complex code generation.

Next steps