Bot TutorialsGrok BotChatGPTClaudeGeminiCopilotPerplexityMore botsGuidesArticlesPapersIndustrialSearch

Industrial control with AI bots: air-gapped Linux tutorials

A practical, Linux-first course on using AI bots around industrial control systems (PLCs, SCADA, HMIs, historians) without making the plant less safe. Written by a bot, checked against public OT security guidance (NIST SP 800-82r3, ISA/IEC 62443), and built for sites that cannot send data to the cloud.

The one-line rule: bots read, explain and draft. Qualified people approve and act. Safety instrumented systems (SIS) and emergency stops never get a bot connection.

Who this is for

Controls engineers, OT/IT security staff, maintenance techs and integrators who want the productivity of an assistant (summaries, troubleshooting help, report drafting, code review of PLC exports) on networks that are segmented or fully air-gapped.

The lessons

1. Ground rules: what a bot may and may not do in OT

The non-negotiables: read-only first, no bot on safety systems, zones and conduits, humans own every write.

2. Run a local LLM on an air-gapped Linux box

Build llama.cpp from source, load a GGUF model, serve it on 127.0.0.1 only. No internet, no cloud keys.

3. Sneakernet: move models and updates across the gap safely

Hash, sign, scan and log every file that crosses the gap. One-way, checked, recorded.

4. Read Modbus TCP data with a bot (read-only)

Poll holding and input registers with Python, then let the local model explain the numbers.

5. Read OPC UA tags securely with a bot

Use SignAndEncrypt, a read-only account and certificate trust lists. Browse, read, never write.

6. Bot-written shift reports and alarm summaries

Turn historian CSVs and alarm journals into a plain-English handover the next shift can trust.

7. Audit logs that a bot cannot rewrite

Append-only logs, hash chains and auditd so every prompt, answer and action is traceable.

8. Human-in-the-loop: approval gates before anything moves

Propose, review, approve, execute, verify. A two-person rule for any setpoint change a bot suggests.

Reference architecture

Keep the bot in its own zone. Data flows up and out of the control network, not down into it.

Level 0-1  Field devices, PLCs, SIS        <- no bot access, ever for SIS
Level 2    HMI / SCADA servers             <- read-only via OPC UA or Modbus gateway
--------   conduit: firewall / data diode, allow-list, logged
Level 3    Site ops DMZ: historian replica + "bot box" (Linux, local LLM)
Level 3.5  Sneakernet station: hash, scan, sign, log every file in or out
Level 4-5  Business / internet             <- air gap: nothing crosses live

Levels follow the Purdue model commonly used with ISA/IEC 62443 zones and conduits. Your site's architecture may differ; map it before you connect anything.

What you need

New to bots? Start with the general guides on catching made-up answers. That skill matters more in a plant than anywhere else.

Questions

Can a chatbot run my plant?

No. Use bots to read data, explain it and draft work. Keep control actions with qualified people and your existing, validated control logic.

Do I need the internet?

No. Every lesson here works fully offline with a local model on Linux.

Where these facts come from

Checked Oct 29, 2026. Tutorials are educational. Follow your site's safety procedures, management-of-change process and the equipment maker's manual. Never test against live production equipment without authorization.

Industrial control with bots: all lessons →

Related: 1. Ground rules: what a bot may and may not do in OT2. Run a local LLM on an air-gapped Linux box3. Sneakernet: move models and updates across the gap safely4. Read Modbus TCP data with a bot (read-only)5. Read OPC UA tags securely with a bot6. Bot-written shift reports and alarm summaries7. Audit logs that a bot cannot rewrite8. Human-in-the-loop: approval gates before anything movesCan AI control a PLC?Run AI offline on LinuxGuidesScripts & templates