How can healthcare teams use private RAG on sensitive data?

Key facts

API surface	OpenAI-compatible /v1/chat/completions; keep your existing integration
Data grounding	Embeddings and RAG are live for policy, coding and knowledge search
Deployment	Plugsky cloud, your VPC, on-prem or air-gapped
Access control	Scoped API keys, rotation and usage analytics; enterprise SSO and RBAC options
Auditability	Request and response logging plus access controls for review
Pricing model	Flat monthly self-serve plans; no per-token billing on self-serve
Free tier	Free plan with 2 free AI models; 14-day full-access trial
Endpoint roadmap	Audio, images, moderation, batch and fine-tuning are coming soon

TL;DR

• Ground answers in approved content with RAG rather than model memory.
• Keep sensitive data inside VPC, on-prem or air-gapped deployments.
• Use scoped keys and audit logs to support privacy and compliance reviews.
• Assign a human owner to every clinical, coding or coverage decision.
• Pilot on policy documents before touching any patient-identifiable data.

How it works, step by step

1. Inventory the content you want searchable: policies, formularies, coding guides and internal procedures.
2. Define a data boundary for the pilot that excludes patient-identifiable data until controls are proven.
3. Choose a deployment target: cloud for public content, VPC, on-prem or air-gapped for restricted data.
4. Build an evaluation set with clinician or coder input so answer quality is judged by domain experts.
5. Ingest, chunk and embed approved documents, and require citations on every answer.
6. Add refusal behavior for questions that fall outside the indexed, approved content.
7. Review logged interactions on a schedule and expand only after accuracy and privacy checks pass.

Try it yourself

Open the RAG sandbox →

Start where decisions are not clinical

Safe first use cases are administrative and informational:

• Policy and procedure search: staff ask questions and get cited answers from current documents.
• Coding and billing support: suggest codes from documentation for certified coders to confirm.
• Formulary and coverage lookup: retrieve plan rules and draft explanations for review.
• Patient FAQ: answer logistics questions such as hours, preparation and documents from approved content.

Each one augments a professional or answers a factual question; none replaces clinical judgment.

A private RAG architecture

The stack mirrors other regulated industries: an ingestion pipeline for approved documents, multilingual embeddings, a vector store inside your environment, and a chat call that answers only from retrieved context. Plugsky embeddings and chat completions are OpenAI-compatible, so the integration is a base-URL change for teams already using OpenAI SDKs.

Air-gapped deployment keeps prompts, documents and vectors inside your network. Use JSON mode when the output feeds a downstream system, and streaming when a person is reading.

Privacy, access and audit controls

Health workloads are governed by privacy rules that vary by jurisdiction, so map your obligations with counsel rather than assuming one framework covers every deployment. Technically, the controls are consistent: minimize and redact identifiers before they reach the model, authorize every retrieval request in your own service layer, scope API keys per application, and retain logs under a defined schedule.

Document what data can leave each boundary, and verify the claim with an architecture diagram before go-live. See what private AI means for the deployment vocabulary.

Human oversight is the safety control

Treat model output as a draft. Require a named human owner for clinical, coding and coverage decisions, display the citations alongside the answer, and make it easy to report a wrong or unsafe response. Track correction rates over time; they are better signals than generic benchmarks.

Honest comparison

Capability	Plugsky	Consumer AI tools	Building in-house
Data boundary	Cloud, VPC, on-prem, air-gapped	Vendor cloud only	You control fully
Grounding	RAG with citations over approved content	Uncontrolled retrieval	You assemble and operate
Access control	Scoped keys, usage analytics, enterprise SSO and RBAC options	Account-level only	Custom identity work
Auditability	Request and response logging	Limited	You build logging
Pricing	Flat monthly self-serve plans; see live pricing	Per-seat	GPU plus operations cost
Time to pilot	Days	Hours, without residency control	Quarters

Frequently asked questions

Can we use patient data with this?

Only after your privacy review approves the deployment and controls. For restricted data, use a VPC, on-prem or air-gapped deployment, minimize identifiers, and keep authorization in your own services.

Is Plugsky certified for health privacy rules?

Plugsky provides controls you can document in your own compliance program, such as scoped keys, logging and private deployment. Confirm current attestations and terms with the Plugsky team and your counsel.

How do we prevent wrong clinical answers?

Restrict the assistant to approved content, require citations, refuse out-of-scope questions, and keep a human decision-maker for every clinical or coverage outcome.

Should we fine-tune on our clinical notes?

Not for a first release. Fine-tuning is coming soon and is better for style than for facts. RAG keeps answers current and traceable, which matters more in a regulated setting.

Can it handle scanned documents?

Vision and document endpoints are coming soon. Today, OCR the scans upstream, embed the extracted text, and link back to the original image for verification.

How do we evaluate answer quality?

Have clinicians or certified coders build a labelled question set, then measure retrieval hit rate, citation correctness and answer accuracy before and after every change.

What about multilingual patient populations?

Multilingual embedding and chat models in the 30+ model catalogue handle mixed-language content. Evaluate on your own languages and terminology.