📡 Idea Radar 每天 8:00 / 18:00 推送 · 18:00 生成精选
reddit · SideProject · 56 分 ← 返回列表

I'm building a memory layer for people who run AI automations for their business. What problems do you hit most?

I'm the founder of Vilix AI, and I'm posting to get feedback, not to sell anything. I run scheduled tasks and agents for my own business, and I kept hitting the same problem: the agents forgot what they had already decided. Rejected options came back, dead ends got retried, and every new run started from zero. So I built a cloud memory layer that connects over MCP. The same memory is available on every tool and device, and the agents check it at the start of a run and save to it at the end.

五维评分

付费意愿
45
需求真实度
55
个人可落地
80
竞争空间
30
新鲜度
72

AI 分析

痛点:AI 智能体每次运行都会忘记此前已经做出的决策,把已否决的方案重新拿出来、把走不通的路再试一遍,而且跨工具、跨设备没有共享的上下文记忆。

目标用户:自己为业务搭建定时任务和 AI 智能体的中小企业主、独立开发者与自动化顾问。

证据:I run scheduled tasks and agents for my own business, and I kept hitting the same problem: the agents forgot what they had already decided. Rejected options came back, dead ends got retried, and every new run started from zero.

变现:按调用量和记忆条数分档的订阅制云端记忆 API(免费额度 + 付费套餐),再加面向团队的自托管和私有部署授权。

MVP:做一个 MCP server:提供 write_decision 和 recall_decisions 两个工具,底层用 SQLite + 向量检索存结构化决策日志(决策、否决理由、时间戳),在每次 run 开始时自动注入相关的历史决策摘要,并支持一键导出和删除;两周内可上线并跑通 3 个真实自动化场景。

竞品:Mem0、Zep、Letta (MemGPT)、Supermemory、LangChain LangMem

摘要:Vilix AI 创始人发布了一个通过 MCP 接入的云端记忆层,用来解决 AI 智能体跨运行时遗忘决策、重复试错的问题,提供免费套餐且数据可导出或删除。该帖是创始人主动征集反馈的发布帖,正文中没有出现付费意愿的原话,也没有收入数据,但相关需求已被 Mem0、Zep、Letta 等多个融资竞品验证。技术实现单人两周可做出 MVP,但赛道拥挤,差异化空间有限。