RedisAI

RedisAI is an open-source Redis module that enables executing artificial intelligence models and managing tensor data directly inside Redis. Developed by Redis Labs, it aims to bring model inference closer to data storage to reduce latency and simplify deployment for applications that combine data management with machine learning.

It introduces new data types for tensors and models and exposes an execution layer that can run models through multiple AI backends. Supported backends include TensorFlow, PyTorch, and ONNX Runtime, with execution on CPUs or NVIDIA GPUs via CUDA. Users store input and output data as Redis keys and reference them in model executions.

RedisAI provides a command set for managing tensors and models and for invoking inference. Typical operations include creating tensors (AI.TENSORSET and AI.TENSORGET), loading models (AI.MODELSET) and executing them (AI.MODELRUN). The system supports multiple models, versioning, and metadata, enabling complex serving workflows and pipelines within Redis.

RedisAI is part of the Redis ecosystem and is maintained as an open-source project with documentation and examples to support deployment in microservices and data-rich environments. It is designed to complement traditional Redis workloads by enabling low-latency AI inference near stored data.