π οΈ The publishing engine
What it is for
The engine separates content from presentation. The notes stay in their own repository, the site code in another: you can write without touching the code, and publish several sites with the same engine.
flowchart LR N["Obsidian notes<br/>and site configuration"] --> M["Engine<br/>Astro + Starlight"] M --> D["Static site"] D --> H["Web hosting"]
What it gives the reader
- Definitions on hover. Hover over the word VRAM: its definition appears without leaving the page.
- Interactive diagrams. Mermaid diagrams can be zoomed, panned and opened full screen.
- Incoming references. Each page lists those that cite it.
- Reading prerequisites. A box shows what to read first.
- French and English. The internal links of a translated page lead to the translated version when it exists.
- Visible freshness. The bottom of the page shows the change and verification dates.
What it gives the author
- The Obsidian way of writing, kept:
[[...]]links, templates, callouts, formulas and Mermaid diagrams. - Configuration on the content side. The title, address, languages and menu are set in the vaultβs
site.config.jsonfile. - Private notes protected. The
_private/folder and excluded files are never published. - Incremental publishing. Only changed files are sent to the server, over FTPS or SFTP.
- Checks. An audit reports broken internal links.
Try it
You need Node.js 22 or later. From a folder next to your Obsidian vault:
git clone https://github.com/DamienBecherini/starlight-obsidian-engine.gitcd starlight-obsidian-enginecp .env.example .env # set VAULT_PATH=../your-vaultnpm installnpm run devThe project README describes the configuration and deployment.
Its limits
- Password protection of the site only works on Apache hosting.
- It expects an organized vault with metadata at the top of each note.
GitHub repository: DamienBecherini/starlight-obsidian-engine, under the 0BSD licence, reusable without conditions.
Very high-bandwidth memory attached to the GPU, storing weights, KV Cache and buffers; its capacity and bandwidth determine which models run without offloading.