I "built" Sana, a vector DB on object storage
Turbopuffer-inspired vector DB with RaBitQ quantization running on plain S3.
Document database with no database server. Tiny, LLM-friendly TypeScript API; DB engine runs to completion in each JS request handler; state lives in S3/R2.
Database engine runs inside each request handler — S3 becomes the DB with conditional creates as commits.
Backend developers, teams building internal tools, serverless architects
DynamoDB · Firebase · PocketBase
Turbopuffer-inspired vector DB with RaBitQ quantization running on plain S3.
FoundationDB layer enabling ACID transactions across documents and key-values with Redis protocol.
Per-actor SQLite eliminates single-table DynamoDB pain—real schema flexibility at scale.
Hierarchical KV store for AI memory, but Electric SQL and PowerSync already solve this.
130× more facts per GiB than vectors, but no semantic similarity matching.
Zero-config single-file persistence plus two query styles (Mongo-like dicts or a Pythonic Q builder) makes everyday prototyping painless. The built-in Studio UI and auto-index hints are thoughtful extras for inspecting data and nudging performance, though this competes with established tiny DBs and SQLite/JSON patterns — great for side projects, but check concurrency and durability needs first.