Key facts
| API surface | OpenAI-compatible /v1/chat/completions; deploy the same calls on-prem |
| Deployment | Cloud, VPC, on-prem and air-gapped with no outbound connectivity |
| Data grounding | Embeddings and RAG are live for drilling, HSE and maintenance documents |
| Long context | Long-context models are available for large technical reports |
| Agents | Function calling is live for historian, maintenance and inventory APIs |
| Access control | Scoped API keys and usage analytics; enterprise SSO and RBAC options |
| Pricing model | Flat monthly self-serve plans; no per-token billing on self-serve |
| Endpoint roadmap | Images, audio, moderation, batch and fine-tuning are coming soon |
TL;DR
- On-prem and air-gapped deployment keeps operational data inside the asset boundary.
- RAG over HSE, drilling and maintenance documents delivers cited answers.
- Long-context models handle large technical reports without fragmentation.
- Agents read historian, inventory and maintenance APIs through your permissions.
- Pilot in the office, then deploy the same API calls at the field site.
How it works, step by step
- Identify the document classes with the highest retrieval value: HSE procedures, well reports, maintenance history.
- Confirm the deployment boundary for each class, including sites with limited or no connectivity.
- Deploy Plugsky on-prem or air-gapped for restricted environments and validate the API gateway locally.
- Ingest and embed approved documents, keeping the vector store inside the same boundary.
- Expose read-only tools for maintenance, inventory and production data with per-user authorization.
- Require citations and escalation paths for any answer that touches safety or regulatory obligations.
- Expand coverage site by site, and log queries to find documentation gaps.
Try it yourself
Open the private LLM deployment estimator →
Field and office use cases
Oil and gas runs on documents that are long, technical and expensive to search:
- HSE procedures: retrieve the current permit, isolation or emergency procedure with the source shown.
- Drilling and completion reports: summarize lessons learned and compare offset wells.
- Maintenance history: trace recurring failures across assets and shifts.
- Regulatory filings: check required fields and deadlines against approved templates.
Each case shortens the time between a question and a verified answer.
Air-gapped architecture for remote sites
Where connectivity is limited or policy forbids external calls, run the whole stack inside the boundary: model serving, retrieval, vector store and logging. Plugsky supports on-prem and air-gapped deployment, and the API stays OpenAI-compatible, so application code developed in the office moves to the field without changes.
Plan updates as a controlled process: model versions and indexes ship as signed artifacts on your own schedule, with evaluations re-run before rollout.
HSE and operational guardrails
Safety content demands strict behavior. The assistant should answer only from approved, version-controlled documents, always show the source revision, and refuse when retrieval confidence is low. Any output that could influence a field decision routes to a qualified person before action.
Log the document version used in each answer, because an answer is only as current as the revision behind it.
From pilot to standard
Start with one site and one document class, measure retrieval accuracy and time saved, then extend to adjacent classes. Keep a registry of approved models and prompts, and review query logs monthly to find where documentation is missing or contradictory. That loop turns a pilot into an operating standard rather than a demo.
Honest comparison
| Capability | Plugsky | Cloud-only AI API | Building in-house |
|---|---|---|---|
| Deployment | Cloud, VPC, on-prem, air-gapped | Vendor cloud only | Your infrastructure |
| Offline operation | Air-gapped mode with no outbound calls | Not possible | Possible with full build |
| Document retrieval | Embeddings and RAG are live | Varies by vendor | You assemble the stack |
| Agent actions | Function calling into internal systems | Limited | Custom orchestration |
| Model choice | 30+ models behind one API | Vendor catalogue | Only what you can host |
| Cost model | Flat monthly self-serve plans; see live pricing | Per-token billing | Hardware plus operations |
Frequently asked questions
Can the model run with no internet connection?
Yes. Air-gapped deployment runs model serving, retrieval and logging inside your network with no outbound connectivity.
How do updates work in an air-gapped environment?
You control the update cadence: ship model and index updates as signed artifacts on your own schedule, and re-run your evaluation set before promoting a version.
Is sensitive production data used for training?
For strict requirements, choose a private or air-gapped deployment so data stays inside your environment. Review the current data-handling terms for cloud plans before rollout.
Which workflows are live today?
Chat, streaming, JSON mode, function calling, embeddings, RAG and agents are live. Images, audio, moderation, files, batch, assistants, responses and fine-tuning are coming soon.
Can it answer from scanned rig documents?
Run OCR upstream with your existing tooling, embed the extracted text, and link each answer back to the original scan for verification. Vision endpoints are coming soon.
How do we evaluate retrieval quality?
Build a set of real field questions with known source documents, then measure whether the correct document appears in the retrieved context and whether the answer stays faithful to it.
How do we start?
Pilot in one office or site using the free plan or a staging deployment, prove accuracy on non-sensitive documents, then move to on-prem or air-gapped for production data.