Replaysafe tackles duplicate actions from AI agent retries

A developer said AI agents can crash mid-task and then repeat the same charge, email, or CRM entry when retried. They open-sourced Replaysafe, a library that wraps non-idempotent calls and uses a fingerprint to check whether the same operation already ran. If it has, the library returns the cached result instead of running the action again. A related post described using a per-session retry budget to stop one flaky tool from causing hundreds of repeated calls and a larger bill.

Key points

  • AI agents can repeat the same real-world action after a crash and retry.
  • Replaysafe creates a fingerprint from the operation type, target, and input.
  • If the same operation was already completed, it returns the cached result instead of running it again.
  • The library says it works with LangGraph, CrewAI, Inngest, n8n, and Airflow.
  • A per-session retry budget can help prevent runaway agent bills from repeated tool calls.

Quick term guide

open-sourced
The code has been made public so others can inspect, use, or contribute to it.
open-source
Software whose code is shared publicly so others can inspect, use, or change it.
non-idempotent calls
Actions that can cause a new effect each time they run, such as charging a card twice.
fingerprint
A compact marker used to tell whether two operations are the same.
cached result
A saved answer from an earlier run that can be reused instead of repeating the work.
retry budget
A fixed limit on how many retries a session is allowed to make.
tool calls
Times when an AI system uses another function, such as search or file access.
open source
Software whose code is available for people to view and often modify.

Sources covering this story (2)

Read original