created a self-hosted AI memory workspace, primarily to solve the problems I often encounter: links, screenshots, chat logs, and temporary text scattered everywhere, making it hard to find when I really need them.

memduck now allows saving this content as personal memory cards, and then using AI to query your own data, with answers trying to reference the original source as much as possible.

Currently available:

- Next.js + SQLite running locally
- Browser extension to capture web pages / selected text
- Telegram running locally
- Entry points for DingTalk, Slack, Discord, Lark, WhatsApp, etc.
- Configuration for OpenAI, Anthropic, Gemini, Ollama, and OpenAI-compatible models
- Q&A and citation based on saved content

project is still early, but the core chain has been running smoothly. It's more suitable for people who want to control their own data, model providers, and channel connections.

GitHub: https://github.com/tageecc/memduck

Screenshot preview: https://raw.githubusercontent.com/tageecc/memduck/main/public/brand/memduck-hero.png