Integration architecture
What connects, through what, and what does not yet.
PHL AI does not talk to every machine. It reads data through formats, gateways and adapters, and this page says exactly which of those exist in the first release.
- Available in MVP
- Designed for the first release and shown in the demo with sample data.
- Requires adapter
- Works through a gateway, adapter or customer configuration on your network.
- Planned
- On the roadmap, designed but not in the first release.
- Future capability
- Requires models, infrastructure or validation that does not exist yet.
01Integration catalogue
Every integration, with its status.
Filter by status or category. “Requires adapter” means it works only through a gateway or converter on your network.
19 of 19 shown
Industrial protocols
- MQTT (via PHL gateway)Requires adapter
Bridge an existing broker or publish from devices through an AWS IoT Greengrass gateway. Outbound TLS only. JSON envelope documented.
- OPC UA (read-only client)Requires adapter
Gateway component subscribes to OPC UA nodes you select. Reads only; no write or method calls are exposed.
- Modbus, EtherNet/IP, PROFINETRequires adapter
Through a third-party protocol converter that exposes OPC UA or MQTT. Not connected directly.
Edge gateways
- AWS IoT Greengrass v2 gatewayRequires adapter
Runs on a customer-approved Linux host. Store-and-forward buffering, certificate identity, schema checks.
- Existing industrial gatewaysRequires adapter
Any gateway that can publish MQTT with TLS client certificates can be configured as a source.
- Edge analyticsPlanned
Deadband filtering and local aggregation on the gateway.
Data imports
- CSV historical importAvailable in MVP
UTF-8 CSV with a header row. Column-mapping wizard, validation preview, row-level error file.
- Parquet historical importAvailable in MVP
Columnar import for large historian exports.
- Sample datasetsAvailable in MVP
Bundled demo cell used throughout this site. Always labelled as sample data.
- Historian connectorsPlanned
Direct historian reads. Today: export to CSV/Parquet and import.
Cloud infrastructure
- AWS IoT Core, Amazon S3, PostgreSQL (Amazon RDS)Planned
Proposed hosting architecture for the platform. Shown as design, not as a running deployment.
- AWS IoT TwinMaker / IoT SiteWiseFuture capability
Under evaluation as optional integrations. PHL AI keeps its own twin model.
Engineering models
- Equipment datasheets (PDF)Available in MVP
Attach to the asset record. No automatic extraction in the first release.
- glTF / GLB 3D modelsPlanned
Optional 3D view bound to asset IDs; the 2D plan view remains the default.
- USD / NVIDIA OmniverseFuture capability
Scene interchange for customers with engineering-grade models and simulation needs.
- External simulation enginesFuture capability
Export scenario inputs and import results with the engine name and validation status attached.
Maintenance systems
- CMMS work history (CSV import)Available in MVP
Import past maintenance events onto the asset timeline.
- Review export for work orders (CSV)Available in MVP
Export review items to create work orders in your CMMS. PHL AI does not create them for you.
- CMMS API connectorsPlanned
Two-way sync with maintenance systems, prioritised with early customers.
02Connectivity architecture
Outbound only. Read only.
A
Physical equipment
- Machines & sensors
- PLCs
- SCADA / historian
- CMMS records
B
Edge connectivity
- OPC UA read-only clientRequires adapter
- MQTT bridgeRequires adapter
- Store-and-forward bufferRequires adapter
C
Data ingestion
- CSV / Parquet importAvailable in MVP
- Validate · normalise units · dedupeAvailable in MVP
- AWS IoT Core + queuePlanned
D
Industrial data platform
- Asset metadata (PostgreSQL)Planned
- Time-partitioned telemetryPlanned
- Files & history (Amazon S3)Planned
E
Digital twins
- Signal map & relationshipsAvailable in MVP
- Versions & freshnessAvailable in MVP
- 3D view (glTF)Planned
F
Analytics & simulation
- Rules & statistical flagsAvailable in MVP
- Scenario models S1 · S2Available in MVP
- Trained predictive modelsFuture capability
G
Application
- Twin explorer & telemetryAvailable in MVP
- Alerts & review queueAvailable in MVP
- ReportsAvailable in MVP
Control boundary. Readings flow from the plant to PHL AI over outbound TLS. There is no write path back: PHL AI does not send commands to PLCs, setpoints or production controls. People act through the systems they already operate.
03Supported data formats
What a file or message must look like.
CSV
UTF-8, header row, one timestamp column (ISO 8601 or a declared format and time zone), one column per signal or a long format with signal and value columns.
timestamp,P301_VIB_DE,P301_BRG_T 2026-10-14T14:10:00Z,4.84,70.8 2026-10-14T14:20:00Z,4.90,71.0
Parquet
Same logical columns as CSV, for large historian exports.
MQTT JSON envelope
One JSON object per reading or batch: gateway, signal key, timestamp, value, unit, quality and sequence number. Published to a per-gateway topic over TLS with a client certificate.
OPC UA
Subscriptions to nodes you select; OPC UA status codes map to quality good / uncertain / bad. Read-only: no writes and no method calls.
04Gateway requirements
What your OT team will be asked to approve.
Host
A Linux host (x86-64 or ARM64) or VM you approve, on a network segment that can reach the OPC UA server or MQTT broker.
Network
Outbound TLS on 8883 or 443 only. No inbound ports to the OT network.
Identity
One X.509 certificate per gateway, revocable and rotatable.
Buffering
Store-and-forward during outages; dropped data is recorded as a gap.
Bring your industrial systems into clearer view.
Start with a focused equipment workflow and expand toward connected digital twins and industrial simulation as your requirements evolve.
Illustrative demo · no signup · not connected to equipment