I often see small projects reach for PostgreSQL because it is “safer for later”. The more useful question: how many concurrent writers will there actually be?
What SQLite does very well
- Fast reads, no network. There is no inter-process latency; the same query is often 5–10 times faster than going over a network connection to a database server.
- Near-zero operations. One file, backups by copying the file. No service that can die at three in the morning.
- Full transactions. SQLite is ACID compliant; with a single writer, behaviour matches a server database.
When to move on
- Very frequent writes from many processes at once (job queues, telemetry).
- Need for cross-region read replicas.
- Dependence on advanced data types available only in specific extensions.
For an internal application with dozens of users, SQLite usually lasts much longer than the initial estimate. Moving later stays possible: if your data access layer isn’t spread across the whole codebase, the migration touches one directory.