Key facts
| API compatibility | OpenAI-compatible /v1/chat/completions; drop-in base URL change |
| Models | 30+ models behind one API; open-weight options for private deployment |
| Typical patterns | Procedure search, maintenance summaries, permit digests, outage comms |
| Integration path | Connects to EAM, GIS and document-management systems through controlled middleware |
| Pricing model | Flat monthly self-serve plans with unlimited fair-use usage; no per-token billing — see the live pricing page |
| Free tier | plugsky-micro and plugsky-lite on the free plan, no card required |
| Deployment | Plugsky cloud, your VPC, on-prem and air-gapped options |
| Compliance posture | SOC 2 Type II and ISO 27001 readiness in progress (not yet certified); validate evidence during diligence |
TL;DR
- Keep AI in the information layer, never in control loops.
- Choose the deployment tier from the consequence of failure.
- Segment AI infrastructure from OT and treat it as untrusted.
- Define degraded-mode and manual fallback for every process.
- Start free with plugsky-micro and plugsky-lite, no card required.
How it works, step by step
- Map candidate use cases to assets and classify by safety and availability consequence.
- Select a deployment tier per use case: cloud, VPC, on-prem or air-gapped.
- Segment AI infrastructure from OT with explicit boundary controls.
- Define degraded-mode behaviour and rehearse it before go-live.
- Implement identity, key custody and audit controls inside your environment.
- Validate on representative operational documents, then expand deliberately.
Original data
Try it yourself
Open the sovereign AI readiness score →
Where an AI API fits in critical infrastructure
The safe pattern is to keep AI in the information layer and out of the control layer. Useful, low-risk starting points:
- Procedure and standards search: surface the current operating procedure or standard for a task with citations.
- Maintenance report summaries: condense inspection and work-order histories into reviewable summaries for engineers.
- Permit and regulatory digests: extract obligations and dates from permits and regulatory correspondence.
- Outage and incident communications: draft status updates for customers and regulators from verified facts.
- Shift handover summaries: structure shift notes into consistent handovers for the incoming team to confirm.
Security, privacy and data handling
Critical infrastructure security starts with segmentation and ends with fallback plans:
- Treat the AI endpoint as untrusted relative to OT; allow no direct path into control networks.
- Deploy private or on-prem when operational data must not leave the site.
- Define a degraded mode and manual fallback for every assisted process.
- Log all requests and keep an inventory of use cases mapped to assets and approvals.
Deployment options and model choice
Operators commonly keep office functions in region-locked cloud while operational documentation runs on-prem or air-gapped. Deployment is a choice among four tiers on the same OpenAI-compatible API: shared cloud, a private endpoint in your VPC, on-prem and air-gapped, with region selection for residency. The catalogue holds 30+ models behind one key, with open-weight models for disconnected sites, and switching provider is a base URL change. Live today: chat, streaming, JSON mode, function calling, embeddings, RAG and agents. Still coming soon: audio, images, moderation, files, batch, fine-tuning, assistants and the responses API. Start on the free plan with plugsky-micro and plugsky-lite, no card, or use the 14-day full-access trial for paid tiers; the live pricing page has current plans.
From pilot to production
The dangerous mistakes are architectural, not textual:
- Connecting the API directly to OT or SCADA networks.
- No manual procedure for when the model or network is unavailable.
- Using shared cloud endpoints for sensitive operational data.
- Treating model version changes as invisible when they alter outputs.
- Scaling a successful office pilot into an operational role without reassessment.
Run a tabletop exercise for AI unavailability before go-live, keep a use-case register with owners, and review every expansion against the consequence of failure. Segmentation, availability and governance outweigh model choice.
Honest comparison
| Capability | Plugsky | Typical per-token API | Building in-house |
|---|---|---|---|
| API compatibility | OpenAI-compatible chat, embeddings and tools | Usually compatible | Full rewrite |
| Deployment | Cloud, VPC, on-prem and air-gapped | Mostly cloud-only | You operate GPUs and serving |
| Data residency | Region selection and sovereign options | Limited regions | You control fully |
| Pricing | Flat monthly self-serve, fair-use usage | Per-token, harder to forecast | GPU plus operations cost |
| Model choice | 30+ models behind one API | Varies by provider | You host every model |
| Industry fit | Procedure search, maintenance summaries, permit digests, outage comms | Generic API, you adapt it | You build every workflow |
Frequently asked questions
Can we keep our existing OpenAI SDK code?
Yes. Plugsky exposes an OpenAI-compatible API, so you change the base URL and model name and keep your code across deployment tiers.
Is there a free plan?
Yes. The free plan includes two free models, plugsky-micro and plugsky-lite, and does not require a credit card.
Should operational data go to shared cloud?
Prefer a private endpoint, on-prem or air-gapped deployment for operational data. Shared cloud suits administrative workloads in a region your policy allows.
What happens when the model is unavailable?
That is what a degraded mode is for: manual procedures, cached outputs or pausing the workflow. Rehearse it, and monitor failover through the status page and audit logs.
Is Plugsky certified?
SOC 2 Type II and ISO 27001 readiness are in progress rather than completed. Track milestones in your risk register and validate compensating controls.
Which endpoints are live today?
Chat completions, streaming, JSON mode, function calling, embeddings, RAG and agents are live. Audio, images, moderation, files, batch, fine-tuning, assistants and the responses API are coming soon.
How do we govern model changes?
Treat model and prompt changes as operational changes: assess impact, test on representative data, record approval, and keep a rollback path.