# Fluxon

An AI-native distributed data plane in Rust that unifies high-performance RPC, KV cache sharing, an elastic message queue and S3-compatible file acceleration behind three standardized interfaces, aimed at AI training and inference bottlenecks.

Fluxon is a distributed data plane built in Rust for AI systems, exposing three standardized interfaces on top of a unified transport and caching foundation. A KV/RPC interface breaks data silos and enables efficient cross-process, cross-node reuse of inference-side KV cache and latent cache; an elastic message queue decouples system dependencies and supports transport across heterogeneous resource pools; and an S3-compatible file, object and cache acceleration system unifies storage forms so one system can cache key-value, file and object data while supporting S3 forwarding and large-scale cross-cluster migration of AI data and model files.

Its own benchmarks, run mainly against RPC, KV and file-system data planes, report that its KV store outperforms MooncakeStore and Redis on read-heavy workloads, and that using a local SSD as a backing layer for DRAM replicas gives several times the hit-payload throughput of comparable Mooncake configurations at larger payload sizes on a single H100 node under SSD pressure.

Benchmark scripts and configuration live under a dedicated test-stack directory in the repository.

Website: https://tele-ai.github.io/Fluxon/