A knowledge management system
README.md

trellis #

trellis is a WIP personal knowledge management (PKM) tool. trellis strives for the following:

  • application speed and efficiency. trellis uses slint and runs at about 120 mb of memory. you can even run trellis with slint's software renderer, for machines that don't have a dedicated gpu.
  • clean ui. i've done a lot of work to make sure that spacing, size, and placement of elements feels just right (disclosure: i'm not a designer). trellis uses adwaita for its main inspiration and almost copies every part of it (without using gtk itself). this is because i personally like the clean look of adwaita compared to qt and it is a well-established design language. it also doesn't strictly adhere to adwaita which means i do have freedom to deviate from it as i see fit.
  • typst for its document format. while markdown is a great plaintext format, it is ill-suited if you want to extend it further. typst is also a plaintext document format, but it is designed in mind for better typesetting and extensibility.
  • security. trellis isn't necessarily a severe-risk application, but it is still made with sandboxing and capabilities in mind. in a proper sandbox, trellis only has access to the vault folder you provide it access to and doesn't need access to your entire filesystem. it also runs purely on wayland and doesn't fallback to x11.

these are the goals i'd like to achieve for trellis:

  • ease of use. trellis tries to ease the incredibly steep learning curve that many PKM tools are notorious first. trellis tries to be as minimal as possible without necessarily compromising on expressiveness. trellis also has a feature (to be named) that allows you to click on any UI element and see the relevant documentation to it.
  • integration with external services. trellis has a concept of "sources". sources can range from consumed media (books, shows, games, music, etc.), research articles, bookmarked sites, tasks/todos, contacts, and so on. ideally these are accessed through solid/atproto, protocols which keep your data saved in a personal server so that other programs and services (such as trellis!) can easily access them on your behalf. this allows you to not be locked in to trellis while still having the capability to make connections between these external sources using your notes.
  • cross-platform support. truth be told, i don't care much about macos/windows/ios. i don't really use these platforms and i'd only ever support trellis if someone else can go through the actual work of making it feel at place on those platforms (not necessarily look, which is different). i do want android support, which means i also want to develop trellis for different screen sizes and input methods in mind.
  • translations. self-explanatory.
  • a11y. i don't have much experience in this myself but i think it would be a shame if i didn't try to develop this for other users in mind. i want trellis to be useful for everyone, regardless of what input devices or methods they use.

building and installing #

this is a rust project. currently this is only for linux. i run and develop trellis purely on nixos, so discretion is advised for other systems.

a justfile (will be) available. even if you don't have just, you can read through it in order to know what commands you need to run. i don't really like putting long instructions in the README on how to compile this (even though i write so much prose in here lol)

setup #

for the necessary libraries, you may look at devenv.nix to know what you need to install on your system.

installing #

after building, you should be mostly done. trellis is just a single binary and should be easy to run. either cargo install --bin trellis --release or moving the binary after the build to your PATH should do it.

flatpak #

included here is a way to automatically create a flatpak build. this only uses cargo so you don't need to install anything else.

this uses flatpak-builder, which builds trellis in a sandboxed runtime which only has the rust toolchain.

this runtime doesn't have access to the internet, so we have to first set up our dependencies by vendoring:

cargo run --bin flatpak-dev setup

before building, there are a few things to note:

  • the flatpak build doesn't use the same build cache as the one you might have in your target/ directory. this is because flatpak-builder wants complete reproducibility to ensure your program comes from one source.
  • however, subsequent flatpak builds will be able to reuse their previous builds. i got annoyed having to constantly build all they way from scratch.

after this is done, you can now start the build;

cargo run --bin flatpak-dev build

when the build is done, you can install it to your user:

cargo run --bin flatpak-dev install

ai disclosure #

yes, i do use ai. i'm not a great or even decent rust programmer. at the same time, if this kind of thing bothers you, i hope you can still give trellis a try. i've put a lot of genuine thought into this and i don't use mainstream closed-source models (i'm too broke for that anyways).

these are the models that i have used, roughly in order of how frequently i use them:

  • deepseek v4.1 flash 1018
  • glm 5.3 flash
  • openrouter/free

llms, even the open weights one, still require massive amounts of resources for training and inference. i am well aware of these and i am not making any statements about ai/agent harnesses being "inevitable" or something along those lines.

i'm also aware that llm-generated code is legally uncopyrightable. i'm not a lawyer so i'm not quite knowledgable. but honestly, if you want to take this code and sell it somewhere else or something, sure. kind of a dick move to be honest, but i'm a believer of the commons.

this section is very all over the place and i don't really want to keep this section this long so i should probably write my thoughts on llms and agent harnesses somewhere soon.

contribution policy #

i'm not really sure yet how this stuff works, this is the first time i've really put a project "out there". i'm mostly building this on my own with my complete vision.

if you encounter any bugs or problems then feel free to use issues.

i don't really have any problem with llm-generated code. but maybe at the very least if you will be discussing with me, then maybe have some decency to write yourself? you can point to what an llm says but yeah.