Records, replays, and compares AI agent sessions to debug non-deterministic behavior and pinpoint divergences.
Sponsored
Agent Replay acts as an MCP server designed to provide deep insight into the execution of AI agents. It captures every action—tool calls, LLM responses, decisions, and errors—along with detailed inputs, outputs, reasoning, and timing. This rich data allows developers to meticulously replay sessions step by step, perform behavioral diffs between runs to detect regressions, and precisely identify where an agent's behavior first diverged from expectations, making it an indispensable tool for debugging and optimizing complex agent systems.
Key Features
010 GitHub stars
02Comprehensive agent session recording
03Session export in JSON or human-readable Markdown format
04Behavioral diffing and divergence detection between sessions
05Detailed action logging with input, output, reasoning, and duration
06Step-by-step agent session replay
Use Cases
01Debugging non-deterministic AI agent behavior
02Identifying behavioral regressions in agent performance
03Analyzing agent decision paths and outputs for optimization
