memcode README

The project's README from github.com/memcode-in/memcode. As one markdown file: README.md.

MemCode logo

End-to-end memory infrastructure for AI.

Give agents, copilots, applications, and workflows persistent memory that survives sessions, context windows, and model changes.

End-to-end Memory React 19 TypeScript
Vite Cloudflare Active Development

## Why Memory

AI systems are powerful inside their current context window, but that context is temporary. Important decisions, user preferences, previous interactions, and learned knowledge disappear between sessions.

A larger prompt is not the same as memory.

Real memory must be:

MemCode is building this layer as infrastructure rather than another temporary context workaround.

Development

Prerequisites

Install dependencies

npm install

Configure the environment

Copy the provided environment template:

cp .env.example .env.local

Update the values for your local or hosted environment. Do not commit production credentials or private keys.

Start development

npm run dev

Create a production build

npm run build

Validation

Run the focused dashboard checks with:

npm run test:byok-usage
npm run test:runtime-settings

The production build also runs TypeScript validation before generating the Vite bundle.

Deployment

The dashboard is configured for deployment through Cloudflare:

npm run build
npx wrangler deploy

Production environment variables should be configured through the deployment platform rather than committed to the repository.

The primary dashboard is available at app.memcode.in.

Project Status

MemCode is under active development. APIs, interfaces, and infrastructure may evolve as we improve retrieval quality, memory evaluation, integrations, and developer experience.

We are building toward a world where every AI system can remember selectively, retrieve reliably, and improve continuously.

Building memory infrastructure for AI, end to end.

Website  •  Dashboard

memcode (memcode-in/memcode)

Searching Database…

这是干什么的?MemCode 是面向 AI 的记忆基础设施:它为智能体、AI 编程助手、应用和工作流提供持久记忆,即使换了会话、超出上下文窗口、甚至换了模型,记忆也不会丢失。
这是我要导入的库,还是一个托管平台?它是一个记忆基础设施层,自带一个仪表盘应用(app.memcode.in),其他智能体、AI 编程助手和应用都连接到它来存取记忆,而不是一个简单的进程内库。
怎么在本地把仪表盘跑起来?克隆仓库,用 npm 安装依赖,复制自带的环境变量模板并填入你自己的值,然后启动开发服务器;生产构建会先做 TypeScript 校验,再用 Vite 打包。
除了 Node.js 和 npm,我还需要准备什么吗?前置条件只需要 Node.js 和 npm,另外要在本地配置好你自己的环境变量——生产环境的凭据或私钥千万不要提交到仓库里。
现在就能直接用到生产环境吗?不行,MemCode 还在积极开发中,随着团队不断改进检索质量、记忆评估、集成和开发者体验,它的 API、接口和底层设施都可能变动。
它存下的记忆,由谁来控制访问权限?MemCode 的设计是通过明确的访问控制和保留策略来管理记忆,而不是让存进去的记忆无限期地开放下去。
与其用这个,为什么不直接加大上下文窗口?更大的提示词只能扩展单次会话能看到的内容;而 MemCode 的记忆能跨会话、跨上下文窗口、甚至跨模型保留,所以智能体在对话结束很久之后仍然记得之前的决定和偏好。
对 MemCode 来说,“选择性”记忆是什么意思?系统不是把所有东西都记录下来,而是设计成自行判断什么值得记住,这样检索就能聚焦在重要的决定和学到的新知识上,而不是一堆噪音。
存进去的记忆怎么取出来?既可以通过语义检索(按意思搜索)查询记忆,也可以通过结构化检索(按字段搜索)查询,而且系统运维人员应该能够直接查看这些记忆。
MemCode 是开源的吗?是的,仓库以 MIT 许可证发布,不过目前使用它的主要方式还是 app.memcode.in 上的托管仪表盘。
Creator
memcode-in
Size
1.9 MB
Stars
603
Updated
CommitSep 08
Age
2mo
Commits · 90 Days