Read the running process.
Live tag values with quality and timestamps, active alarms, historical data, device status, and server logs come back as structured data. Your agent sees the process, not only the project.
Thermite SCADA
An AI-native SCADA and HMI platform. Tags, screens, and alarm and history settings live in plain files that your own AI agent can read, change, and publish as a draft for your team to review.
[[tags]] name = "bowl_temp" path = "filler/bowl_temp" data_type = "f64" device = "filler-plc" address = "ns=2;s=Filler.BowlTemp" eng_units = "°C" [tags.history] enabled = true
The idea
Tags, alarm and history settings, operator screens, and queries are ordinary text: TOML, HTML, CSS, JavaScript, and SQL. There is no graphical designer and no configuration database to export from. A Git checkout is the working project, so every project change is a diff your team can read, and your agent can work on the system the way an engineer does.
Live tag values with quality and timestamps, active alarms, historical data, device status, and server logs come back as structured data. Your agent sees the process, not only the project.
One command checks a change and publishes it as a draft at its own address. Review it there, then publish: the running gateway picks up supported project changes without a restart.
With authentication enabled, write permissions are set per role and start switched off. A write refused by those permissions stops before it reaches the device, and the audit log records both writes and refusals.
AI-native by design
Thermite SCADA has no built-in AI assistant, by design. Your agent works with the project files, Git, the thermite command line, and the running system’s HTTP and WebSocket interfaces: the same working surfaces your engineers use. No plugin, skill, or MCP server is needed to begin. You choose the agent and the model.
Want to see it? Bring your own agent to a live session and watch it build a working SCADA application against simulated devices.
Ask for a live demoYouLine 2’s filler stopped four times on the night shift. What happened?
AgentI’ll check the alarm journal, then the history around each stop.
Read projects/line-2/tags/filler.toml9 tags, history on each; low-level alarm on bowl_level
Read logs/alarms.jsonl4 × filler/bowl_level:lo since 22:00, each cleared within 6 min
Fetch /api/logs?level=warnno device or connection warnings in the log
Fetch /p/line-2/api/history/filler/bowl_level?start=-10h&mode=aggregated&interval=1mlevel falls to its low limit four times
Fetch /p/line-2/api/history/filler/infeed_pressure?start=-10h&mode=aggregated&interval=1mpressure sags about 10 min before each alarm
AgentFound it. Each stop follows the same pattern: infeed pressure sags from 2.4 to 1.6 bar, the bowl drains, and the low-level alarm comes in about ten minutes later. The filler is being starved, not faulting.
AgentThis was read-only. I changed nothing in the project or on the line.
YouThat matches the wash-downs on line 1. I’ll take it to maintenance.
Connectivity
Six native protocol adapters connect field data to tags, screens, alarms, and history.
Need another protocol or device workflow? Talk with us about your equipment, network, and data requirements.
Discuss your protocol needsUnder the surface
line-2/
├── project.toml # name, roles, devices in scope
├── tags/
│ ├── filler.toml # tags, alarms, history
│ └── capper.toml
├── screens/
│ ├── overview/
│ │ └── index.html
│ └── filler/
│ ├── index.html
│ ├── screen.js
│ └── style.css
├── transactions/
│ └── batch-log.toml # log each batch to SQL
└── queries/
└── batch-report.toml # a named SQL queryFair questions
No, and that is deliberate. Instead of adding an assistant to the product, we made the product itself workable by the agent you already use: its files, its command line, and its live interfaces.
Any agent that can read and edit files, run commands, and make network requests. Nothing Thermite-specific is installed in the agent. How well a given model handles controls work is yours to evaluate, and we can help with that.
Start with reading. Project edits are diffs your team reviews, and a draft can be tried before publishing to production. With authentication enabled, write permissions start switched off and writes are audited. Thermite keeps changes reviewable; who may publish stays your decision.
No. Engineers work with the same files, commands, and interfaces directly; screens are ordinary HTML, CSS, and JavaScript written by hand or with whatever tools they prefer. An agent is a way to move faster on a system your team understands, not a requirement for running one.
It keeps running. A licensed installation never phones home and its licence does not expire. Your project and its history stay in your own repository, as plain text in standard formats. Thermite SCADA is proprietary software; where you need more assurance, a source-code escrow agreement can be arranged.
Your next step