Architecture Overview

Haltless is a multi-tenant platform built for industrial predictive maintenance. This page explains how the major components fit together.

System architecture

                    ┌─────────────────────────────────────────┐
                    │              Haltless Cloud              │
                    │                                         │
  Edge Agents ─────▶│  POST /ingest  ──▶ TimescaleDB          │
  (HTTP push)       │       │              (time-series)      │
                    │       ▼                                 │
  OPC-UA / Modbus ─▶│  Direct Ingestion ──▶ PostgreSQL        │
  (server polling)  │  Background Task      (relational)      │
                    │       │                                 │
                    │       ▼                                 │
                    │  Anomaly Detection ──▶ Alert Engine      │
                    │  EWMA + Rate-of-Change  │               │
                    │       │                 ▼               │
                    │       │          Escalation Policies     │
                    │       │          ──▶ Webhooks / Email    │
                    │       ▼                                 │
                    │  RUL Prediction (AI) ──▶ Dashboard       │
                    │                                         │
                    │  REST API (FastAPI) ◀── Client Portal   │
                    │  WebSocket /ws/dashboard                │
                    └─────────────────────────────────────────┘

Components

Edge agent

A lightweight agent that runs on-premises (or on an edge gateway). It collects sensor data from your machines and pushes it to the Haltless cloud via HTTP. See Edge Agent Overview.

Authentication: X-API-Key header.

Direct ingestion

For OPC-UA and Modbus TCP devices, Haltless can poll them directly from the cloud , no local agent required. You configure an Ingest Source with the device address, and a background task handles polling. See Direct Ingestion Guide.

REST API

The core interface for all operations. 120+ endpoints organized into resource groups:

Group	Purpose
Machines	Register, update, query machine fleet
Sensor Data	Time-series ingestion and retrieval
Alerts	Threshold alerts, acknowledgment, snooze
Predictions	AI-powered RUL and anomaly detection
Maintenance	Events, work orders, compliance sign-offs
Fleet	Cross-machine status and KPIs
Reports	PDF/CSV export
Settings	API keys, permissions, AI config

WebSocket

Real-time streaming via wss://api.haltless.io/api/v1/ws/dashboard. After connecting, the client sends its JWT as the first text message to authenticate. The server then streams live updates for the tenant's machines.

Database layer

• PostgreSQL , relational data (machines, users, tenants, work orders, etc.)
• TimescaleDB , time-series hypertables for sensor readings with automatic chunk management
• Redis , token blocklist, login throttle, session management

Anomaly detection pipeline

Haltless uses a multi-layered detection approach:

1. Static baselines , per-metric mean/stddev computed over a 7-day window (min 30 samples)
2. EWMA (Exponentially Weighted Moving Average) , adaptive baselines that track gradual drift
3. Rate-of-change , velocity analysis to detect sudden spikes in sensor values
4. Composite rules , AND/OR logic combining multiple metric conditions

When anomalies are detected, alerts are generated and routed through escalation policies to notification channels (email, Slack webhooks, Microsoft Teams, push notifications).

Multi-tenancy

Every resource is scoped to a tenant (organization). Tenants are isolated at the database level , all queries include a tenant_id filter. Users belong to exactly one tenant and have one of three roles:

Role	Description
admin	Full access. Manages users, settings, billing, API keys.
operator	Can manage machines, maintenance, work orders. Cannot modify settings or users.
viewer	Read-only access to dashboards, machines, and alerts.

Subscription tiers

Rate limits, machine counts, and data retention are enforced per-tenant based on their subscription tier:

	Free	Pro	Enterprise
Max machines	2	50	10,000
Data retention	7 days	365 days	3,650 days
API writes/min	60	300	1,200
API reads/min	300	1,500	6,000