A basis for what you and your AI know.

kbasis is an open knowledge base on a graph whose links have types. Your AI searches it, stores in it and helps you understand it. You run it, you own it, and any AI that speaks MCP can use it.

Run it yourself

Try it on your own machine with Docker, at localhost:3000. To reach it from another device, or from an AI elsewhere, give it a hostname with HTTPS. The README shows how.

terminal
$ git clone https://github.com/kbasis/kbasis
$ cd kbasis
$ sh deploy/env.sh
$ docker compose -f deploy/compose.yaml up -d --build --wait
# open the setup link env.sh printed, and create a passkey

What kbasis promises

  1. It stays MIT.Never relicensed to something more closed.
  2. Nothing is withheld.A hosted version never gets a capability the open one lacks.
  3. Your knowledge stays yours.Open, readable formats you can take anywhere.
  4. Any AI is your choice.Any client, any model, your own key.
  5. It's built in the open.Decisions and their reasons are public.

Roadmap

  • Now
    Document editingYou write in the web app; your AI reads and edits the very same blocks as markdown, from any client that speaks MCP. Neither of you works on a copy, so an edit from either side lands in the one document. Text, links and pasting already work, and arranging blocks on a canvas is coming.
  • Next
    Typed linksToday a link points at a document, or at one block inside it. Next, every link says how two things relate (one leads to the other, contains it, describes it, or just sits near it), and you can name that relation in your own words. Whatever a link points to shows everything that points back, so you and your AI can follow a thought in either direction.

From the build

The first post is coming.