Quản lý cuộc sống trên máy chủ của bạn,
với hệ thống mà AI thực sự dùng được

tududi is open source under the MIT licence. You can run it, modify it, and build on it without paying anyone. There are no seat counts, no usage limits, and no features held back for a paid tier: the self-hosted version is the whole product. The paid options on this page sell hosting and support, not access.

Yes, and you do not need to buy anything first. The MIT licence permits commercial use, including inside a business and inside a commercial product, with no revenue threshold and nothing to report. If your organisation needs a support commitment or an invoice for its procurement process, that is what the Business Licence is for. It buys priority support, a response commitment, and paperwork, not permission.

Yes. It is built primarily by one maintainer, with contributions from the community, and it ships often. Every release is listed on the GitHub releases page, linked from the Changelog in the menu above, so you can judge the pace yourself rather than take our word for it. Development happens in the open: issues, a public roadmap, and a Discord where questions get answered.

The app is MIT licensed and the source is public, so it cannot be withdrawn from you. Anyone can keep running it, and you can fork it. For Managed instances, your data lives in a standard tududi database that is not in a proprietary format; if the Managed service ever shut down, you would get your data and a guide to moving it onto your own server.

Docker and somewhere to run it. You pull the image, run the container, and mount a directory for the database. It is comfortable on a small VPS or a machine on your own network, and the exact commands are in the Self-Host section above and in the docs.

Everything lives in a single SQLite database file inside the volume you mounted. Backing it up is copying that file somewhere safe, and restoring it is copying it back. tududi also snapshots the database automatically before it applies migrations, keeping recent copies. There is no proprietary format and no export step standing between you and your own data.

Yes. A single instance can host multiple users, and projects can be shared with read-only or read-write access, which the tasks and notes inside them inherit. Self-registration is off by default, so an admin either creates accounts or deliberately opens sign-ups. Nobody gets in because you forgot to close a door.

tududi installs as a PWA rather than a native app. Add it to your home screen on Android or iOS, or install it on a desktop, and it behaves like an app: it reads from cache when you are offline and syncs your changes when you reconnect. There is nothing to download from the App Store or Play Store.

Yes, and that is the normal way to run it. Set TUDUDI_TRUST_PROXY so the app trusts the forwarded headers from nginx, Caddy, or Traefik. If you skip it, logins and IP logging behave strangely in ways that are hard to diagnose, so the app raises a specific error rather than failing quietly. The accepted values are documented in the README.

No. Both AI layers are off until you switch them on, and tududi is a complete system without either. If you never enable them, nothing is missing and nothing nags you.

That depends which of the two layers you mean, and the difference matters. The MCP server runs inside your own instance: an agent such as Claude Desktop connects to your server, so the connection and your data stay where you put them. The AI Assistant is different. It calls an OpenAI-compatible provider, so if you point it at a hosted one, the contents of those requests go to that provider. Point it at a local model instead, such as Ollama or LM Studio, and nothing leaves your machine. It stays off until you supply a key, so this is always your choice rather than a default.

MCP is the Model Context Protocol, an open standard for connecting AI agents to real systems. tududi ships an MCP server exposing 59 tools spanning tasks, projects, areas, goals, habits, notes, people, tags, views, inbox, search and metrics, so an agent can capture, find, update and finish work rather than just talk about it. It is opt-in, and it works with Claude Desktop, Cursor, Windsurf, Cline, or any other MCP client.

You are paying to not run it. Managed gives you a container and database isolated from every other user, updates applied for you, and support with a response commitment. If you already run Docker and a reverse proxy, self-hosting is genuinely the better deal and we would rather you did that. Managed exists for people who want the system without becoming its sysadmin.

Instances that go unused for about half an hour are paused, and start again by themselves the next time you open yours. Usually that shows up as a slightly slower first load. Occasionally it takes long enough that you will see a short "waking up" page and can retry. Your data is untouched either way: pausing stops the container, it does not delete anything.