Active Development — v0.6.0 shipped, v0.7.0 in progress

The distributed database
that does more

Redis-compatible. Adaptive tiered storage (memory → SSD → S3). Built-in feature flags, rate limiting, config management, and semantic caching for AI workloads — all in one system.

Get Started View on GitHub
# Works with any Redis client — zero migration
docker run -p 6379:6379 kiradb/kiradb:latest

redis-cli SET hello world
+OK
redis-cli GET hello
"world"
Why KiraDB
One system. No external dependencies.
🔄

Redis-Compatible Protocol

RESP3 wire protocol — every Redis client (Jedis, Lettuce, redis-cli) works without any code changes.

💾

Adaptive Tiered Storage

Hot data lives in memory, warm data on NVMe SSD, cold data auto-archived to S3. Zero manual configuration.

🏳

Built-in Feature Flags

Percentage rollouts, instant kill switch, per-user targeting — without LaunchDarkly or a separate service.

⏱

Distributed Rate Limiter

Token bucket and sliding window algorithms. Enforced consistently across all nodes via CRDT counters.

🧠

Semantic Cache

Cache LLM responses by meaning, not exact match. "capital of France?" hits the same entry as "what is France's capital?" Saves 60–80% on LLM costs.

🔗

Raft Consensus

Strong consistency across a 3-node cluster. Leader failure recovers in under 500ms. No data loss.

Built With
Engineering choices that matter
Layer Technology Why
Language Java 25 Virtual threads, records, sealed classes, value types
Networking Netty 4.x Non-blocking I/O — same choice as Kafka and Cassandra
Storage LSM Tree + RocksDB Hand-rolled LSM, RocksDB JNI as production alternative
Consensus Raft Understandable distributed consensus, from scratch
Cold storage AWS S3 SDK v2 Auto-archival of cold SSTables, async prefetch on access
Vector search Weaviate ANN search for semantic cache embeddings
Build Gradle multi-module Clean module boundaries, independent builds
Roadmap
Where we are and where we're going
v0.1.0
RESP3 Protocol Server GET, SET, DEL, PING, TTL, EXPIRE — any Redis client works out of the box
Done
v0.2.0
Storage Engine WAL, MemTable, SSTables, LSM Tree, Bloom Filters, Compaction
Done
v0.3.0
Raft Consensus 3-node cluster, leader election, log replication, failure recovery
Done
v0.4.0
Adaptive Tiered Storage MemCache (hot, 35% of heap) → LsmStorageEngine (warm, SSD), pluggable orchestrator
Done
v0.5.0
CRDTs GCounter, PNCounter, LWWRegister, MVRegister, ORSet — with serialization & convergence tests
Done
v0.6.0
Built-in Services FLAG.* (LWW + sticky bucketing), RL.* (sliding-window over GCounter), CFG.* (history + Netty server-push)
Done
v0.7.0
Semantic Cache Vector embeddings, cosine similarity, ANN search for LLM prompt caching
In Progress
v0.8.0
Dashboard + Java SDK React dashboard for real-time cluster state; fluent Java SDK with connection pooling
v1.0.0
Production Ready Benchmarks, full documentation, hardening pass against the Phase 13 backlog