Welcome to KalamDB
KalamDB is a SQL-first, real-time database that scales to millions of concurrent users through a revolutionary table-per-user architecture. Built in Rust with Apache Arrow and DataFusion, it combines the familiarity of SQL with the performance needed for modern chat applications and AI assistants.
Why KalamDB?
Traditionally, building real-time applications requires a stack of different technologies:
- A relational database for storage
- Redis/Kafka for real-time events
- A backend API to glue things together
- A WebSocket reverse proxy/server
- Constant fine-tuning, caching, scaling, sharding, backpressure handling…
That setup becomes complex, expensive, and fragile.
KalamDB replaces this entire stack with a single, unified database that:
- Isolates data per user for scale and privacy
- Pushes live changes instantly via WebSocket subscriptions
- Is GDPR-friendly by design — deleting a user removes their entire storage directory
- Avoids unnecessary complexity — one binary, one connection string
- Scales to millions with minimal tuning
Key Highlights
| Feature | Description |
|---|---|
| ⚡ Sub-millisecond writes | RocksDB hot tier for blazing fast buffered writes |
| 📡 Live SQL subscriptions | Real-time data over WebSockets |
| 🧍 Per-user isolation | Each user gets their own table & storage |
| 💾 Hot + Cold storage | RocksDB for writes, Parquet for analytics |
| 🌍 Multi-cloud storage | Local, S3, Azure, GCS backends |
| 🔒 RBAC & Auth | Built-in role-based access control |
| 🦀 Built in Rust | Performance, safety, and concurrency |
Current Version
Latest release: v0.3.0-alpha2
What’s the name about?
Kalam (كلام) means “speech” or “conversation” in Arabic — fitting for a database designed specifically for storing and streaming human conversations and AI interactions.
Get Started
Head to Getting Started to run KalamDB in minutes with Docker, or jump straight to the SQL Reference to see the query language.