About
Lucania is my project: a Claude.ai-style interface for models that run on your own machine. Here is why I wrote it, how I work and what is inside.
I'm Raven. I remember what matters: here is who makes Lucania, and why.
00 / Motto
I build sites that hold up.
I take a project from start to finish and keep it going over time. Lucania is one of them.
01 / Why Lucania
Ollama runs language models on your machine, or on a server you manage. The open question is how to talk to them.
I wanted the Claude.ai interface: the conversation in the center, a sidebar that files your history, answers that write themselves as they come. With one rule from the start: nothing goes to the cloud.
So I wrote Lucania. It plugs into Ollama locally, or into a remote server through WireGuard. It keeps your conversations in a SQLite database on your disk and installs without administrator rights.
The need
Models that run at home
The shape
A Claude.ai-style interface
The rule
Nothing in the cloud
02 / How I work
Five steps, in this order. A project does not end the day it goes live.
Before I open an editor, I ask the questions: who uses it, for what, on which machine. For Lucania the answer fit in one line: chat with Ollama models, no cloud, in an interface as clear as Claude.ai.
A dark background, rounded surfaces, a tight typeface and mono for the details. Red is a signal and nothing else. I also drew the logo: 100 proposals, one kept.
Astro 7, vanilla JavaScript, Tailwind CSS v4 and daisyUI 5. SQLite for the conversations, Tauri v2 for the desktop app, Ollama for the models. No JavaScript framework.
Two files, Lucania-Setup.exe and Lucania-Update.exe, built by a GitHub workflow and published under names that never change. This site is hosted on GitHub Pages.
11 versions in 2 days, from v0.1.3 to v0.1.13. An installer that updates without touching your conversations. Feedback comes in through GitHub issues.
03 / The story of the wolf
For the logo, I set up a lab: 100 proposals, split into 10 series from A to J. I compared them side by side, then kept a single one, H2: a low-poly wolf.
It is cut from seven facets, triangles of different opacity. Nine other animals from the same family followed, and each one carries a Lucania feature. You meet them in the headers of this site.
Kept · H2
A few cousins
04 / Under the hood
Ollama for the models, a memory of facts, a web stack with no JavaScript framework.
The engine
Ollama runs the language models, on your machine or on a server you manage. Lucania is its chat interface: you install Ollama, download the models you like, and Lucania connects to it.
Memory
In Settings > Memory, you pick a model and manage its facts. They are given to it at the start of every conversation, and everything stays on your computer.
Facts are filed per model. A fact is 500 characters at most.
At the start of every conversation, the model receives its facts.
The model can suggest facts to remember from your conversations. You also add, edit and delete facts yourself.
Memory is switched on per model and per conversation.
The more facts the model remembers, the more its answers fit you: your projects, your tools, your preferences. Nothing is sent elsewhere to get there.
Memory builds on Obsidian: your facts stay readable notes, kept at home.
What the app repository contains, according to its README.
The base of the interface, server-rendered with the Node adapter.
All the browser code, with no JavaScript framework.
The styles, written as utility classes.
The interface components, built on Tailwind.
The local database of your conversations, in a file on your disk.
The shell of the desktop app for Windows.
The VPN tunnel you manage to reach a remote Ollama server.
05 / The project in numbers
11
published versions
Git tags of the app
13
themes
picked in Settings
2
languages
French and English
100
logo proposals
in 10 series, from A to J
License
MIT
The code is open source.
06 / Versions
Each pill is a Git tag of the app repository. It opens the version page on GitHub.