EXPLAINABLE PREDICTIVE MAINTENANCE

A Python edge agent reads your existing PLCs and historians over OPC UA, Modbus TCP, Modbus RTU, Siemens S7, EtherNet/IP, Mitsubishi MELSEC, Omron FINS, Beckhoff ADS, BACnet/IP, MQTT, Sparkplug B, AVEVA PI Web API, CSV file watch, or JSON file watch. Three deterministic anomaly detectors score every machine with a published formula, and a multi-tenant audit chain keeps every action traceable for seven years.

Edge agent on your network

The Python edge agent runs on any Linux box next to your machines. Configure each ingest source over OPC UA, Modbus TCP, Modbus RTU, Siemens S7, EtherNet/IP, Mitsubishi MELSEC, Omron FINS, Beckhoff ADS, BACnet/IP, MQTT, Sparkplug B, AVEVA PI Web API, CSV file watch, or JSON file watch. Telemetry is buffered locally in SQLite, so the agent keeps collecting even if the WAN drops for days and auto-flushes when connectivity returns. No new sensors, no proprietary hardware.

Works through WAN outages

The edge agent keeps a local SQLite buffer. Telemetry continues to be collected through outages that last days or weeks. When connectivity returns, buffered readings auto-flush to the cloud in order, and a schema-version check clears stale data after upgrades so silent corruption is not possible.

Deterministic anomaly detection

Three independent detectors run in parallel against every reading: a z-score against a rolling baseline, an EWMA that adapts to gradual drift, and a rate-of-change check that catches velocity anomalies even when the absolute value is still in range. The 0 to 100 health score follows a published deduction table, not a black-box model, so every score traces back to specific events.

CMMS workflow and ten notification channels

Anomalies above threshold can open a CMMS work order in the same platform, with parts pulled from inventory, an assigned operator, and a digitally signed completion record. Ten outbound notification channels keep the rest of your stack in the loop: Webhook, Email, Slack, SMS, Microsoft Teams, PagerDuty, Opsgenie, Telegram, ServiceNow, and Jira.

BUILT TO SCALE

From a 10-machine pilot to a 10,000-machine fleet

Every number here is the actual platform capability you can put into a contract. No marketing approximations, no asterisks, no "results may vary".

14Input protocols (OPC UA, Modbus TCP, Modbus RTU, Siemens S7, EtherNet/IP, MELSEC, FINS, ADS, BACnet/IP, MQTT, Sparkplug B, PI Web API, CSV, JSON)

3Deterministic anomaly detectors

10Outbound notification channels

5UI languages out of the box

3ERP adapters (SAP, Oracle, Dynamics)

7 yearsHMAC-SHA256 audit chain retention

10,000Machines per tenant at full scale

< 100 msWebSocket push from sensor to browser

Sub-100 ms real-time fleet view

Per-tenant WebSocket channels push readings to the browser in under 100 ms from sensor to screen. JWT-authenticated, scoped to your tenant, with connection caps that keep a noisy neighbour from impacting your fleet.

Tamper-evident audit grade

Every audit log row carries a monotonic sequence number and an HMAC-SHA256 hash chained to the previous row. Append-only database triggers block deletes without an explicit retention flag. PostgreSQL Row-Level Security pins every query to the current tenant, and the audit chain is retained for seven years to support SOC 2 CC2.2 and CC4.1.

Predictable per-machine billing

Per-machine pricing tailored to your fleet size. Contracts run annually as standard and are invoiced monthly or annually, whichever suits your finance team, with the term and frequency set out in your contract. Any applicable VAT or sales tax is shown before you confirm.

International from day one

Five UI languages out of the box: English, German, Spanish, Hungarian, and Chinese. Country names are localised at render time, and tenant-level single sign-on over OpenID Connect with PKCE connects to your corporate identity provider. US-hosted by default.

DOCUMENTATION

Setup, configuration, and the full deployment reference

Edge-agent installation, per-protocol setup for all fourteen inputs, system requirements, and the full deployment reference live on the docs site.

Open docs.haltless.io

COMMON QUESTIONS

Questions about how the platform works

How does the health score get calculated?

health_score = (1 − w) × alert_score + w × anomaly_score, where w is a configurable anomaly weight per tenant. Every deduction traces back to a specific event through a published per-event table, so a score is reproducible on paper by your reliability engineer or your auditor.

What happens when the plant loses its internet connection?

The edge agent buffers to a local SQLite store and keeps collecting through WAN outages lasting days or weeks, then ships the backlog when the link returns. Traffic is outbound only from the agent to the cloud, over TLS 1.3 with optional mutual TLS.

Can Haltless write to my PLCs or change a setpoint?

No. The agent reads and never writes. There is no control action, no setpoint change and no path back into the process, which is what keeps it out of IEC 61511 safety scope and out of your control-system validation envelope.

How many machines can one deployment handle?

The same platform runs a ten-machine pilot and a ten-thousand-machine fleet. What changes between them is the number of edge agents and the per-machine licence, not the architecture.