01
Connect the equipment you actually have
MQTT, OPC-UA, SQL, Modbus, Open Protocol, IPC-CFX, IPC-HERMES-9852, REST, watched folders, cameras and serial machines over TCP bridges.
STANDARD + LEGACY SOURCESThe broker gap
A UNS solves the plumbing. It does not solve the work.
The broker is online. Topics are structured. Data moves cleanly between systems. Then the project stalls, because a raw topic tree still means little to an operator, quality engineer or maintenance team.
The usual patch is another custom layer: scripts, dashboards and reporting projects that become technical debt as soon as they ship.
Available data and usable data are different problems. UNS Navigator closes that gap without replacing the systems already running your plant.
Product overview
See the working layer in action.
The walkthrough follows the full path: industrial sources, a controlled visual canvas, plant topology, explainable detection, operator dashboards and secure AI-callable outputs.
1Build and test against realistic simulators before site access.
2Give telemetry topology, asset meaning and cited operational knowledge.
3Publish the same governed result to operators, REST clients and AI assistants.
One working surface
The page is the model, the dashboard and the AI context.
Place sources and functional blocks on a canvas, then wire them together. The same structure drives what people see, how data moves and exactly what an agent can reason over.
02
Make the structure explicit
Typed asset models, process flowcharts, relationships and curated knowledge turn flat values into a contextual model of the operation.
CANVAS + TOPOLOGY + KNOWLEDGE03
Ship results, not another project
Generate operator apps, expose stable REST endpoints and publish selected pipeline results as tightly scoped MCP tools.
HUMAN + MACHINE READABLEContext by construction
AI that only sees the evidence you selected.
An agent can read a block only when it is enabled, configured and inside its scope. Reports list the blocks used and cite the source behind factual claims.
✓No hidden whole-namespace embedding or private copy of plant data.
✓Durable page memory, reusable skills and bounded multi-step investigations remain visible.
✓Use Anthropic, OpenAI, Azure, OpenRouter, compatible endpoints or local Ollama inference.
Telemetry needs topology
Turn the topic tree into a working model of the plant.
A raw MQTT hierarchy tells you where a value lives. It does not explain what feeds what, where a process branches or which asset and procedure give that value meaning.
✓View the namespace as a tree or a live knowledge graph.
✓Bind telemetry to process shapes, asset models and typed relationships.
✓Feed cited SOP and manual knowledge into a human-reviewed proposal queue.
Explainable detection
Catch the problem and show why it fired.
Detection reports the method, field, value and threshold behind every verdict. Forecasting projects when a trend will cross a target, and refuses to invent an answer when the signal cannot support one.
RANGEZ-SCOREMADRATEFLATLINEEWMA
Build before site access
Prototype the solution without a live factory.
Generate realistic JSON telemetry locally, inject faults, record a stream once and replay it at your desk. Build the pipeline and operator surface before touching a production broker.
✓Sine, ramp, drift, random-walk, sawtooth and injected spike patterns.
✓Clearly badged simulated data cannot be mistaken for production telemetry.
✓Walk into the site meeting with a working prototype instead of a slide.
Beyond the dashboard
Make plant results callable.
A screen is useful to a person. A governed data product is useful to everything else. Any pipeline result can serve the same stored output through REST and, with one controlled setting, as an MCP tool an authorised AI assistant can call.
RESTStable read-only endpoint
→
MCPScoped tool for AI agents
A complete working layer
More than a dashboard over MQTT.
UNS Navigator carries context from the live signal to the operator, the historian, the knowledge behind it and the systems that consume the result.
Reusable industrial asset models
Define an asset contract once, bind fields to mixed sources and reuse the same apps and detectors across instances. Historise selected sources in TimescaleDB with retention, compression and rollups.
Reviewable plant knowledge
Add typed relationships and operator notes, then turn SOPs, work instructions and manuals into cited proposals. A person accepts every addition.
Generated live surfaces
Generate dashboards from curated pages, collect them in an App Hub and run a fullscreen rotation for control-room displays.
Check the work, not only the machine
Capture station images, define a known-good view and turn visual checks into ordinary telemetry verdicts that can alarm, trend and historise.
Use the identity your plant already has
Local accounts, Windows / Active Directory and Microsoft Entra sign-in, with server-enforced viewer permissions plus separate audit and activity logs.
Typed, versioned industrial pipelines
Run manually, on a schedule, from a stream or from events. Dry-run traces each step and stubs sinks before anything leaves the workspace.
MQTT · OPC-UA · SQL · MODBUS-TCP · OPEN PROTOCOL · IPC-CFX · IPC-HERMES-9852 · REST · SERIAL/TCP · FILES · CAMERAS · MCP
Trust by construction
Safe enough to place beside live production.
UNS Navigator is additive. Plant-facing connectors read existing systems, while deliberate egress remains narrow, named, deny-by-default and auditable.
01
No PLC or equipment command pathSource connectors do not write to PLCs or source SQL databases and do not directly command equipment.
02
Sensitive connectors fail closedSerial bridges, watched folders and outbound REST remain off until an operator explicitly allow-lists a target.
03
Named, controlled write pathsMQTT publish, historian storage, notifications and supported human-approved tools are separate, deliberate capabilities rather than hidden side effects.
04
Credentials stay server-sideSecrets are stripped from responses, pages, exports and logs. Blocks retain a connection reference, not the credential.
05
Generated code runs isolatedGenerated apps and user transforms run inside constrained execution boundaries rather than inheriting unrestricted plant access.
Local-first deployment
Runs on your hardware. Genuinely.
Deploy the Navigator container with optional TimescaleDB history and FalkorDB graph services. Licensing uses offline signature verification and does not require a cloud account or phone-home connection.
Deployment profile
Containerised and air-gap ready
Verified on x86-64 and ARM64, including compact cabinet-level hardware. Local AI through Ollama keeps inference inside the facility when required.
DOCKERARM64X86-64OLLAMAOFFLINE LICENCE
$ docker compose up -d
✓ navigator healthy
✓ data_dir mounted
✓ licence trial · 30 days
GET /health
{
"status": "ok",
"network": "local",
"phoneHome": false
}
✓ navigator healthy
✓ data_dir mounted
✓ licence trial · 30 days
GET /health
{
"status": "ok",
"network": "local",
"phoneHome": false
}
Start today, decide later
A full 30-day trial starts on first boot.
No licence key, account or internet connection is needed to evaluate the full product. When the trial ends, the workspace becomes read-only instead of deleting data or disabling deployed apps.
30-DAY OFFLINE TRIAL · WORKSPACE REMAINS VIEWABLE AND EXPORTABLE
Questions
Clear boundaries.
Is UNS Navigator an MQTT broker?
No. It connects to your broker. The built-in simulator can also generate realistic industrial streams when no broker or plant connection is available.
Does it replace SCADA, MES or a historian?
No. It reads and connects them, then provides the working layer above. Selected sources can also use the optional TimescaleDB historian.
Will it write to PLCs or source databases?
No. Plant-facing source connectors do not write to PLCs or source SQL databases, and the product has no direct equipment command path. Deliberate egress paths such as MQTT publish, notifications and supported approved tools are separate, narrow and controlled.
What controls what the AI can see?
A block must be enabled, configured and inside the agent's scope. Each gate can narrow access but cannot widen it. The resulting report lists the blocks it used.
Can it work before we connect to a live factory?
Yes. Simulators generate realistic streams without a broker, and recorded streams can be replayed for development, demonstrations and operator training.
Can it learn from SOPs and work instructions?
Yes. Documents can produce proposed relationships and notes with page-and-sentence citations. A person reviews each proposal before it enters the plant knowledge graph.
What is MCP support for?
An exposed pipeline result can become a read-only MCP tool, allowing an authorised AI assistant to query a curated plant result without receiving direct access to its underlying broker or database.
Does it need internet access?
No. It can run and validate its licence offline. Local inference through Ollama is also supported for deployments where plant data must remain inside the facility.
What hardware does it need?
A Docker-capable host. UNS Navigator is verified on x86-64 and ARM64, including compact hardware suitable for line-level deployment.