I know the purpose of sqlite, however, considering the needs of isolated tenants, with small volumes, I'd love to have:
- server manages thousands/millions of individual SQLite database files, one per tenant, behind a single gRPC/HTTP API
- provides per-tenant write serialization, a connection pool
- WAL-based replication to object storage (S3-compatible) - optional
- and an optional extension system for encryption (SQLCipher)
- optional extension vector search (sqlite-vec)
It will be suitable for products with tens of thousands of small, isolated tenant datasets where cost, operational simplicity, and data isolation matter more than global transactions or cross-tenant queries.
I'd probably use Go for this.
Note: Turso already has libsql with a server component but seems they are are fully committed to rewriting their own database.
This is just an idea, looking to see what's your view on this. It will be open-source if launched.
Wrote some specs here with help from Claude
https://docs.google.com/document/d/1xgbPemWHatuCppw2x0_fgUV9...
Use case:
- you need physical segregation of data per tenant
- you want to have all the benefits of Sqlite ecosystem
For example:
- HIPAA with each tenant separated
- one failing tenant does not affect all the others.
