📡 Idea Radar 每天 8:00 / 18:00 推送 · 18:00 生成精选
reddit · SideProject · 48 分 ← 返回列表

I built an open-source tool for replaying and debugging AI agent executions. Looking for contributors and feedback.

I've been working on StepFork , an open-source project that helps developers understand what happened during an AI agent run, reproduce failures offline, and generate regression tests. The goal is to make debugging agent workflows more reliable and less dependent on reproducing issues in a live environment. What StepFork supports Record AI agent executions and inspect traces. Replay recorded executions to investigate failures. Fork executions to experiment with fixes. Minimize

五维评分

付费意愿
30
需求真实度
45
个人可落地
82
竞争空间
28
新鲜度
72

AI 分析

痛点:AI Agent 的执行过程是黑盒,一旦失败,很难在本地稳定复现和定位,也缺少能防止问题再次出现的回归测试。

目标用户:构建 LLM Agent 与工作流的独立开发者及中小团队,尤其是 LangGraph、LangChain 的使用者。

证据:The goal is to make debugging agent workflows more reliable and less dependent on reproducing issues in a live environment. / I'm now looking to grow the contributor community and make the project useful to more developers. / Recording AI agent executions and inspect traces. Replay recorded executions to investigate failures. Fork executions to experiment with fixes. (全文没有任何付费或收入表述)

变现:采用开源核心加托管云服务的方式:把 trace 存储、团队共享回放和 CI 回归测试做成订阅;企业版支持自托管与 SSO 授权。

MVP:只面向 LangGraph 或 LangChain 做轻量 CLI:录制一次 agent 运行的 trace,离线回放并 diff 两次执行的差异,自动把失败 trace 最小化为最小可复现脚本,并生成 pytest 回归用例。用 pip 安装,并将其接入 GitHub Actions,以一条命令接入。

竞品:LangSmith、Langfuse、AgentOps、Braintrust、Arize Phoenix、W&B Weave

摘要:作者发布了开源工具 StepFork v0.2.1,用于录制、回放、fork 和最小化 AI Agent 执行轨迹,并生成回归测试;同时公开招募贡献者,列出了 Windows/macOS CI 冒烟测试、CLI 错误处理测试覆盖、LangGraph 集成教程等具体 issue。帖子属于开源项目推广,没有提到任何付费意愿、收入或用户规模数据,而且该赛道已有 LangSmith、Langfuse、AgentOps 等成熟竞品。