How PixelCore is built and hosted.
The engineering view of the platform: how payloads are decoded, where objects, state and history are stored, which interfaces applications and integrations use — and whether the whole thing runs in the cloud or inside your own perimeter.
For the engineers who will run it
Signals flow left to right; each stage has one well-defined job, and every application reads the same core.
Sources
Sensors, meters, gateways, building and facility systems
Bridges
Protocol adapters and drivers decode payloads into typed properties
PixelCore
PostgreSQL · TimescaleDB — objects, state and full history
Dispatcher
Rules, derived metrics, alarm state, RPC registry
API layer
GraphQL queries and subscriptions · gRPC between services
Applications
Monitor · Board · Tables · Assistant · Admin
Decoding
Raw payloads are parsed once, in the bridge layer, into typed properties with units — so no application repeats the work.
Storage
PostgreSQL holds the object model; TimescaleDB holds the time series. Current value and history are queried together.
Interfaces
A typed GraphQL layer serves apps, integrations and the assistant agent; the RPC registry carries commands back out.
Run it where your data needs to live
The same platform, the same model and the same applications in both cases — the choice is about where the data sits and which network the integrations reach.
Cloud
Managed and quick to provision — the usual starting point for a pilot and for multi-site rollouts.
- Provisioned in days rather than months
- Scaling storage and compute as the estate grows
- Regional hosting options, confirmed per project
Customer-hosted
The same platform inside your own network and security perimeter, for data-residency and OT requirements.
- Data stays within your infrastructure
- Suited to regulated and segmented environments
- Integrates with on-site building and plant systems
What it takes to run it
One containerised stack
The same build runs in our cloud and on your own hosts.
Outbound only
Bridges connect out over TLS — no inbound ports towards the OT network.
Roles, not shared logins
Users, permissions and device credentials live in PixelAdmin; SSO per project.
Policy per property
High-rate signals never force the whole estate onto one retention policy.
Standard database tooling
PostgreSQL and TimescaleDB backups — model and history restore together.
The platform watches itself
Gateway health and last-seen times are properties, so silence is alarmable.
Host sizing depends on point count and sample interval rather than floor area — the estimator below gives the order of magnitude, and we confirm the specification against your own data before a pilot.
Estimate the scale of your deployment
Six inputs are enough to see the data volume the platform would carry.
Your estate
Rough numbers are fine. Volume assumes one stored sample per property per interval at roughly 24 bytes compressed, before downsampling.