reductstore/reductstore

What it solves

ReductStore provides a unified system for storing and querying multimodal time-series data, specifically for robotics and industrial IoT. It eliminates the need to combine multiple separate systems—such as a time-series database (TSDB), generic object storage, and metadata indexes—to manage binary-first data like camera frames, sensor readings, and ROS bags.

How it works

It functions as a high-performance, time-indexed object storage system. Every record is stored in its native binary format and organized by timestamp. Users can attach labels (metadata) to these records, allowing them to retrieve specific data based on a combination of time ranges and operational context (labels).

Who it’s for

• Robotics platforms that need to replay telemetry, logs, and camera frames around specific incidents or model failures.
• Industrial IoT pipelines that collect binary payloads at the edge and need to forward only filtered, relevant data upstream to the cloud.
• Edge AI systems requiring historical retrieval, labeling, and data retention policies without building a custom storage stack.

Highlights

• Multimodal Support: Stores images, video, sensor readings, logs, and ROS bags in native binary formats.
• Contextual Querying: Retrieve data using time ranges and labels (e.g., "all images from camera-1 where status is 'error'").
• Efficient Replication: Replicate only selected records to the cloud to save bandwidth and and storage costs.
• Lifecycle Management: Integrated quotas and lifecycle policies to automate data compression and deletion.