Industry Solutions

How do you build AI agents for critical infrastructure?

Critical infrastructure agents summarise maintenance history, retrieve procedures, support incident documentation and assemble compliance evidence. They must sit on the IT side with no path to control systems, use read-only tools, and leave every operational action to authorised staff. Plugsky provides the compatible API with VPC, on-prem and air-gapped deployment options.

Key facts

Agent stackChat completions, function calling and embeddings are live
Use casesMaintenance history summaries, procedure retrieval, incident documentation, compliance evidence
OT boundaryKeep agents on the IT side with no write path to control systems
ToolsRead-only access to maintenance, document and case systems
DeploymentPlugsky cloud, your VPC, on-prem and air-gapped options
Models30+ models; small aliases for retrieval, stronger ones for synthesis
Pricing modelFlat monthly self-serve plans with unlimited fair-use usage
RoadmapBatch and assistants endpoints are coming soon

TL;DR

  • Keep the agent on the IT side with read-only tools only.
  • Ground procedure and maintenance answers in current revisions.
  • Leave every operational action to authorised staff and existing change control.
  • Run air-gapped where policy forbids external traffic.
  • Measure retrieval accuracy and documentation time, not actions taken.

How it works, step by step

  1. Pick one IT-side workflow such as procedure retrieval or compliance evidence assembly.
  2. Confirm the architecture keeps no write path to OT, SCADA or protection systems.
  3. Index approved procedures, maintenance history and regulatory documents with revision metadata.
  4. Expose read-only tools for document, work-order and case lookups.
  5. Require citations and route any operational decision through existing change control.
  6. Log tool calls and model versions, then measure accuracy and documentation time saved.
1Pick one IT-sideworkflow such asprocedure retrieval2Confirm thearchitecture keepsno write path to3Index approvedprocedures,maintenance history4Expose read-onlytools for document,work-order and case5Require citationsand route anyoperational6Log tool calls andmodel versions,then measure

Try it yourself

Open the AI agent comparison →

Use cases on the IT side of the line

Critical infrastructure operators get value from knowledge work that never touches control systems:

  • Maintenance knowledge: summarise asset history and retrieve the correct procedure for a work order.
  • Incident documentation: assemble timelines and evidence from logs and notes for after-action review.
  • Compliance evidence: collect and organise records for audits and regulatory reporting.
  • Procedure Q&A: answer staff questions from current, approved documents with citations.

Operational actions continue through existing change-control and approval processes.

Architecture with an OT hard boundary

Draw the boundary first and build inward:

  • No write path: agents cannot send commands or modify control-system state, by design and by network policy.
  • Read-only tools: document search, work-order lookup and case retrieval.
  • Revision-aware index: store revision and effective date with every chunk and check freshness before answering.
  • Composer: outputs with citations and explicit assumptions.
  • Human authorisation: engineers and operators remain accountable for every action.

Log tool calls and model versions so incidents and audits can be reconstructed.

Isolation, resilience and records

Many operators require air-gapped environments, and some require that AI workloads never share infrastructure with operational systems. Plugsky supports cloud, VPC, on-prem and air-gapped deployment with the same compatible API, so the agent code, prompts and evaluations move intact. Design for degraded operation too: if the agent is unavailable, the underlying procedure and change-control process must still function. Keep logs for the period your assurance framework requires, and treat any safety or security discrepancy as an escalation rather than an answer.

Piloting and measuring

Prototype on plugsky-micro or plugsky-lite with one asset class and approved documents, then use the 14-day full-access trial to test a stronger model on multi-document synthesis. Measure retrieval accuracy, citation correctness, stale-revision rate and documentation minutes saved per work order. Self-serve plans are flat monthly with fair-use usage, which suits steady operational workloads and avoids storm-season billing spikes. Expand by asset class or site, with versioned prompts and a re-run of the evaluation set after each document revision.

Honest comparison

CapabilityPlugskyTypical agent frameworkBuilding in-house
OT separationAgents stay on the IT side with read-only toolsDepends on your designYou design the boundary
Tool callingLive function calling over internal document and case APIsFramework-specific adaptersYou build and maintain
DeploymentCloud, VPC, on-prem and air-gappedUsually cloud-onlyYou operate GPUs and serving
PricingFlat monthly self-serve, fair-use usageVaries by framework and modelGPU plus operations cost
Model choice30+ models behind one APIVaries by providerYou host every model

Frequently asked questions

Can the agent control equipment?

No. Keep agents on the IT side with read-only tools and no write path to OT or control systems; operational actions stay with authorised staff.

Which endpoints do infrastructure agents use?

Chat completions, function calling, JSON mode and embeddings are live. Batch and assistants endpoints are coming soon.

Can we run it air-gapped?

Yes. Air-gapped deployment is available, and the compatible API means the same agent code runs there without modification.

Is there a free plan?

Yes. The free plan includes plugsky-micro and plugsky-lite with no credit card, enough to prototype procedure retrieval and evidence assembly.

How do we handle document revisions?

Store revision metadata with each indexed chunk, require the citation and verify freshness before returning an answer.

How does pricing work?

Self-serve plans are flat monthly with unlimited fair-use usage. See the live pricing page for current plans and enterprise options.

How should we measure it?

Track retrieval accuracy, citation correctness, stale-revision rate and documentation minutes saved per work order.