RobustMQ
Next-Generation AI Data Communication Infrastructure
A unified, high-performance, low-cost communication substrate for data flowing between AI training clusters, millions of Agents, IoT devices, and the cloud.
Deeply Optimized for Three Scenarios
Not a generic MQ stack — each scenario is purpose-designed
AI Training Acceleration
Eliminate GPU Idle Time
- S3/MinIO direct connect + 3-tier cache
- No data pre-import, eliminates I/O bottleneck
- Million-scale Topics for Agent isolation
IoT Full Pipeline
Edge to Cloud Unified
- MQTT in / Kafka out, one system replaces two
- Tiny memory footprint for edge gateways
- Offline cache + auto-sync full pipeline
Big Data Pipeline
Kafka Protocol Compatible
- Full Kafka protocol compat, zero migration
- 4-mode storage engine, per-Topic config
- Auto-tier cold data to S3, 95% cost reduction
Three Components, Clear Boundaries
Compute, storage, and coordination fully separated — each layer scales independently, delivered as a single binary
Designed from the Core
Extreme Performance
Rust built, microsecond latency, zero GC pauses, million QPS
Dual Protocol
Full MQTT 3.1/3.1.1/5.0 & Kafka compat, unified storage
Pluggable Storage
Memory/RocksDB/FileSegment/S3, per-Topic granularity
Zero Dependencies
Single binary, built-in Raft consensus, no ZooKeeper
Compute-Storage Sep.
Stateless Broker scales horizontally, storage evolves independently
ISR Consistency
ISR replication, Active/Sealed Segment tiering, scale without migration
Start a Cluster in 3 Steps
Single binary, zero external dependencies, one command per node