Industrial IoT & Smart Factory

10KHz sensors tell the plant what is happening. Agent memory helps the plant remember what worked last time. ZeptoDB combines edge time-series ingestion with maintenance context, cache lookup, and replayable agent decisions.

Why manufacturers choose ZeptoDB

10KHz+ ingestion per sensor

Vibration, current, pressure, and temperature at the rate the physics demand. Lock-free pipeline with no jitter on the hot path.

Real-time analytics in SQL

Rolling averages, SPC, anomaly detection, threshold alerts — standard SQL window functions, no UDFs to maintain.

Edge to cloud

ARM Graviton on the line, multi-node cluster in the plant, Parquet HDB on S3 for long-tail history. Same binary, same SQL.

Industrial protocols

OPC-UA, MQTT, Telegraf, and Kinesis connectors for PLC, SCADA, brownfield sensor buses, and cloud streams. OPC-UA covers scalar, array, string, structured, history, alarms, browse, and server-mode surfaces.

Maintenance memory

Keep prior diagnoses, operator notes, work orders, model calls, and tool outputs searchable by machine, session, and incident.

Where it fits

Predictive maintenance

Ingest vibration, temperature, current, and oil-debris sensors at 10KHz+
Rolling statistics (EMA, stddev, percentiles) for drift and early-warning signatures
ASOF JOIN to correlate sensor anomalies with maintenance events and work orders
Use OPC-UA arrays, structured fields, strings, Historical Access backfills, Telegraf line-protocol input, Kinesis streams, or MessagePack batches through the same quality and routing policy

Quality control & SPC

Real-time statistical process control with window functions
Detect out-of-spec measurements within microseconds of the sample hitting the bus
Historical analysis from Parquet HDB for root-cause investigation and yield review

Energy & OEE

Sub-second power consumption aggregation across line, cell, and facility
xbar time bucketing for shift, hourly, and daily OEE and energy-per-unit reports
Peak-demand detection and load balancing analytics in standard SQL

Digital twin & process optimization

High-fidelity historian for simulation calibration and what-if analysis
Join live sensor streams with setpoints and batch records
Feed production ML models with the same zero-copy path used in research
Expose selected ZeptoDB symbols back to PLC / SCADA clients through OPC-UA server mode when open62541 server support is enabled

Industrial agent workflows

Retrieve similar failures before recommending a maintenance action
Pair live alarms with prior work orders, operator notes, and pinned safety procedures
Cache repeated explanations for common machine states
Store agent decisions and tool calls as AgentOps telemetry for audit and improvement

Example: vibration anomaly detection

-- Detect anomalous vibration readings using rolling statistics
SELECT
    sensor_id,
    ts,
    vibration,
    ema(vibration, 100) AS ema_100,
    vibration - ema(vibration, 100) AS deviation
FROM sensor_readings
WHERE
    ts > now() - interval '5 minutes'
    AND abs(vibration - ema(vibration, 100)) > 3 * stddev(vibration)
        OVER (PARTITION BY sensor_id ORDER BY ts ROWS 1000 PRECEDING)
ORDER BY deviation DESC

Deployment options

Environment	Configuration
Line / cell edge	Single node, ARM Graviton, 16 GB RAM, local NVMe
Plant cluster	3-node cluster, 10 GbE, Parquet HDB on NAS
Cross-site historian	Multi-node EKS, Parquet HDB on S3, cross-site replication
Hybrid	Edge for real-time, cloud sync for cross-plant analytics and ML

See Production Deployment for reference architectures.

For connector details, see Integrations, Telegraf Output, HTTP MessagePack ingest, C++ OPC-UA reference, and Kinesis feed handlers.

Why this matters

Industrial data platforms typically stitch together a historian, a stream processor, a vector store, a feature store, and a warehouse. ZeptoDB replaces the hot path with one model for live facts and agent context, from a ruggedized edge node to the cloud. Fewer integrations, less bespoke glue, and a clearer path from pilot line to production agent.

Share plant, edge, or connector feedback through the public project: GitHub Discussions → · Issues →