Architecture

Slite runs a client-only Nuxt 4 frontend served by a single Nitro Node.js backend process, storing all data locally without external cloud dependencies.

What happens when someone opens a short link

1. Resolution: A visitor requests a short link on your domain. Public redirect resolution executes before administrative authentication.
2. Lookup & cache: Slite checks its in-memory link cache. If missing, it fetches the record from authoritative SQLite storage and populates the cache.
3. Evaluation: Slite evaluates stored link rules: expiration, password protection, the persisted unsafe flag (DoH checks occur at link creation or editing time, never during redirects), and smart routing (device or country overrides).
4. Redirect & logging: Slite returns the HTTP redirect (default 301) and asynchronously records a visit event into local DuckDB analytics.
5. Administration: You manage links, export data, and inspect reports through /dashboard or the REST API, both authenticated by NUXT_SITE_TOKEN.

Local storage

All persistent state lives inside the local directory specified by NUXT_DATA_DIR (defaults to /data):

Relative path	Storage engine	Required?	Description
slite.sqlite	SQLite / Drizzle	Yes	Authoritative link store, tags, and core application settings
analytics.duckdb	DuckDB	Recommended	Visit events, click metrics, and geographic query logs
files/images	unstorage fs	Optional	Uploaded social preview images
backups	unstorage fs	Optional	Automatic daily link JSON backups and manual exports

SQLite and the in-memory cache

SQLite is the single authoritative source of truth for all links and application state.

The link cache uses the unstorage memory driver:

• It resides entirely within the memory of the active Node.js process.
• It is process-local, rebuildable on demand, and never acts as a secondary authoritative store.
• When a process starts, the cache is empty and populates lazily from SQLite.
• Committed link mutations synchronously invalidate corresponding cached entries.
• If a cache operation encounters an issue, caching is safely disabled and lookups fall back directly to SQLite.

Uploaded images and link JSON backups are written via the unstorage filesystem driver under files/images and backups.

Process model and constraints

Slite is designed for a single-process architecture:

• Run only one container and one Node.js process per persistent data directory.
• SQLite, DuckDB, and in-memory caches are not distributed systems. Clustered process managers, horizontal replicas, and shared network filesystems (NFS, SMB) are unsupported.
• Container replacement or upgrades are safe as long as the underlying local volume is retained and previous processes have stopped before new ones launch.

Optional services

• AI assistance: Uses xsai to connect with any OpenAI-compatible provider (OpenAI, DeepSeek, Ollama, etc.). Requests submit only destination URLs without fetching page content.
• Click webhooks: Asynchronously posts signed JSON event payloads to an external webhook endpoint upon link clicks.
• GeoIP metadata: Resolves geographic dimensions using local MMDB databases. Release Docker images bundle the DB-IP City Lite database; lookups fail open if no database is present.
• Near-realtime dashboard: Employs a 10-second analytics polling loop paired with a client-side queue that replays events at roughly one per second; it does not rely on SSE or WebSocket connections.