reddit · SideProject · 09-27 18:50 44 分 ← 返回列表

I built a tiny C++ AI agent that runs 24/7 on cloud and you can connect to it via web

I've been working on this for a while and finally want to make it public The idea is basically: instead of setting up a VPS, installing an agent, configuring models, MCPs, skills, keeping it alive, etc., you just create an agent and give it a job. It runs 24/7 on its own server. I'm targeting: 4GB RAM 2 vCPUs ~€10/month Bring your preferred AI model Real server deployments Hooks / MCPs / Skills / Rules Agent configures itself based on what you ask it to do

五维评分

付费意愿

需求真实度

个人可落地

竞争空间

新鲜度

AI 分析

痛点：自建 24/7 常驻 AI Agent 需要买 VPS、装环境、配模型/MCP/技能并保证不掉线，而常见 JavaScript Agent 运行时吃 1-2GB 内存，小机器根本跑不动。

目标用户：想跑常驻自动化 Agent（远程写代码、挂 Minecraft 服务器、持续监控任务）的独立开发者与小型技术团队。

证据：The agent runtime itself is written in C++ and currently uses around 30MB RAM to avoid the 1-2GB RAM overhead of standard JavaScript agents. / I'm targeting: 4GB RAM, 2 vCPUs, ~€10/month / It's not launched yet. I'm trying to figure out whether there's enough real interest to turn this into a company.

变现：托管订阅（约 €10/月/实例，含 4GB RAM + 2 vCPU 与常驻运行），加上一次性或开源版的自托管授权（跑在 Mac mini / 树莓派上）。

MVP：两周版：做一个极轻量的常驻 Agent 模板，Docker 一键部署到 Fly.io/Railway，网页端提供终端与文件浏览、任务循环、单一 LLM Provider（OpenAI/Anthropic）接入与基础 MCP 调用，按实例收费并做等待名单转化。

竞品：E2B、OpenHands、Manus、Devin、Fly.io Machines、Railway、LangChain/LangGraph

摘要：作者用 C++ 写了一个仅约 30MB 内存的 AI Agent 运行时，可 24/7 跑在云端，目标配置 4GB RAM/2 vCPU、约 €10/月，主打免去 VPS 与 MCP/模型配置、并规避 JS Agent 的 1-2GB 内存开销。项目尚未上线，只有 demo 和 waitlist，作者自己也在问『是否真有足够兴趣做成公司』，没有收入或付费用户数据。机会方向成立（轻量常驻 Agent 托管），但需求与付费意愿均未被验证。