展示HN：Memograph CLI - 一款用于诊断AI代理“内存故障”的工具
嗨，HN，
我一直在构建AI代理和助手，但遇到了一个令人沮丧的问题：它们不会大声失败，而是悄悄地忘记事情。
用户需要重新解释偏好设置，代理会与之前的回答相矛盾，且上下文在没有明确原因的情况下重置。
我开发了Memograph CLI作为调试工具，以分析对话记录并显示：
- 代理忘记了什么
- 连续性在哪里中断
- 矛盾和重复的上下文
- 由于重新提示而导致的估计令牌浪费
它可以在本地运行，支持纯文本或JSON格式的记录。
示例：
```
$ memograph
```
输出：
```
认知漂移评分：41/100
遗忘的偏好：3
令牌浪费：29%
破坏信任的矛盾：1
```
目标不是替代你的代理框架，而是让开发者能够看到内存失败的情况。
代码库：[https://github.com/memographAI/Memograph-CLI](https://github.com/memographAI/Memograph-CLI)
非常希望能收到反馈，特别是来自正在生产环境中构建代理的人的意见。
查看原文
Hi HN,<p>I’ve been building AI agents and copilots, and kept running into a frustrating problem: they don’t fail loudly, they forget things quietly.<p>Users re-explain preferences, agents contradict earlier responses, and context resets without any clear visibility into why.<p>I built Memograph CLI as a debugging tool to analyze conversation transcripts and show:<p>- what the agent forgot
- where continuity broke
- contradictions and repeated context
- estimated token waste due to re-prompting<p>It works locally and supports plain text or JSON transcripts.<p>Example:<p>$ memograph<p>Output:<p>Cognitive Drift Score: 41/100
Forgotten preferences: 3
Token waste: 29%
Trust-breaking contradictions: 1<p>The goal isn’t to replace your agent framework, but to give developers visibility into memory failures.<p>Repo: <a href="https://github.com/memographAI/Memograph-CLI" rel="nofollow">https://github.com/memographAI/Memograph-CLI</a><p>Would love feedback, especially from people building agents in production.