Key facts
| Access control | Scoped API keys with rotation; enterprise SSO and RBAC options |
| Deployment | Cloud, VPC, on-prem or air-gapped with region selection |
| Auditability | Request, model and revision-level response logging |
| Data grounding | Embeddings and RAG are live for manuals, service bulletins and records |
| Structured output | JSON mode returns inspection and compliance records in fixed schemas |
| Models | 30+ models including long-context options for technical documentation |
| Pricing model | Flat monthly self-serve plans; no per-token billing on self-serve |
| Endpoint roadmap | Images, audio, files, batch and fine-tuning are coming soon |
TL;DR
- Answer only from approved revisions and record which revision was used.
- Keep export-controlled and airworthiness documentation inside the boundary.
- Require qualified human review for maintenance and design outputs.
- Give each system and vendor integration its own scoped key.
- Log enough to reconstruct any answer for an airworthiness review.
How it works, step by step
- Inventory document classes: maintenance records, service bulletins, design data, compliance files.
- Identify export-controlled or restricted content and confine it to on-prem or air-gapped deployment.
- Issue per-application keys with rotation, and connect enterprise SSO and RBAC for console access.
- Index approved documents with revision metadata, and require answers to cite the revision.
- Set review gates so maintenance and design outputs are approved by qualified personnel before use.
- Log request IDs, model versions, retrieved revisions and outputs under your retention policy.
- Run periodic audits against a sample of answers to confirm revision accuracy.
Try it yourself
Open the AI API key security checklist →
Airworthiness and revision control
Aviation documentation changes constantly, and an answer from a superseded revision is worse than no answer. Store revision metadata with every indexed document, retrieve the current approved version by default, and make the assistant cite the revision it used. When a manual or bulletin is updated, re-index before the new revision reaches the floor.
JSON mode helps downstream systems: inspection and compliance records can be returned in schemas that your records system validates and stores.
Export control and document boundaries
Some technical documentation is subject to export-control or partner-confidentiality restrictions that dictate where it may be processed. Keep those collections in on-prem or air-gapped deployments with no outbound connectivity, and keep their retrieval indexes separate from general service documentation. The API surface stays the same, so the application does not change when the boundary does.
Keys, systems and vendor access
Maintenance, engineering and planning systems each have different owners and risk levels. Issue separate keys per system and environment, store them in a secrets manager, and revoke promptly when a contract or role ends. Vendor and MRO partner integrations get their own credentials with logging, so partner activity is attributable. Avoid shared keys spanning engineering and commercial workloads.
Traceability and human review
For airworthiness reviews, the question is not only what the assistant said, but what it read and who acted on it. Log request ID, model and version, retrieved revisions, output and the following human decision. Require qualified review for anything that touches maintenance instructions, repair data or design changes, and keep the assistant in a retrieval-and-summarize role rather than a decision role.
Honest comparison
| Control area | Plugsky capability | Common gap | Owner |
|---|---|---|---|
| Revision control | RAG with revision metadata and citations | Answers from superseded manuals | Technical publications |
| Export-controlled data | On-prem and air-gapped deployment options | Restricted docs in shared tools | Compliance |
| Identity | Scoped keys, rotation, enterprise SSO and RBAC options | Shared engineering credentials | Security engineering |
| Audit trail | Request, model and revision logging | No reconstruction path | Quality assurance |
| Vendor access | Per-integration keys and usage analytics | Unmanaged MRO credentials | Procurement |
| Human review | Citations plus structured output for sign-off | AI output used directly | Chief engineer office |
Frequently asked questions
How do we guarantee the current revision is used?
Store revision metadata with every indexed document, re-index on updates, and make citations include the revision. Audit a sample of answers regularly to confirm it.
Can export-controlled documentation be processed?
Only in deployments that meet the control requirements, typically on-prem or air-gapped with no outbound connectivity. Keep those indexes separate and document the boundary.
Should AI write maintenance instructions?
No. It should retrieve and summarize approved documents for a qualified person who reviews and signs off. Keep decisions with accountable engineers and inspectors.
What should we log for airworthiness reviews?
Request ID, model and version, retrieved document revisions, output and the human decision that followed. That supports reconstruction during audits or investigations.
Is scanned technical documentation supported?
Vision and file endpoints are coming soon. Today, OCR the documents upstream, embed the extracted text with revision metadata, and link back to the original scan.
How do we handle partner or MRO access?
Issue distinct keys per partner integration with their own logging, limit retrieval to the documentation each agreement covers, and review access quarterly.
Where should a pilot start?
Use public manuals and internal service documentation on the free plan, validate retrieval and citation accuracy, then move restricted collections to a private deployment.