Key facts
| Access control | Scoped API keys with rotation; enterprise SSO and RBAC options |
| Deployment | Cloud, VPC, on-prem or air-gapped for sensitive investigations |
| Auditability | Request, model and response logging for editorial review trails |
| Data grounding | Embeddings and RAG are live for archives, transcripts and rights metadata |
| Structured output | JSON mode returns tagging and rights records in fixed schemas |
| Models | 30+ models behind one OpenAI-compatible API |
| Pricing model | Flat monthly self-serve plans; no per-token billing on self-serve |
| Endpoint roadmap | Audio, images, moderation, files and fine-tuning are coming soon |
TL;DR
- Treat unpublished editorial and source material as your most sensitive class.
- Separate investigation workspaces from general newsroom archives.
- Carry rights and provenance metadata through every automated step.
- Log requests, sources and editors for anything that reaches publication.
- Keep editorial judgement with named journalists and editors.
How it works, step by step
- Inventory AI use cases across research, transcription, tagging, archives and audience work.
- Classify content by publication status, source sensitivity and rights status.
- Choose deployment per class: cloud, VPC, on-prem or air-gapped.
- Issue per-desk and per-workflow keys with rotation and central inventory.
- Define log fields and retention: request ID, model, sources, output, reviewing editor.
- Keep provenance and rights metadata attached as content moves through pipelines.
- Require editorial review before publication and named approval for legal-sensitive work.
Try it yourself
Open the embedding model comparison →
Editorial pipelines and data classification
Media content divides into published archives, working drafts, unpublished investigations and source material. Each class needs its own rule, and source protection usually demands the tightest controls because disclosure can harm individuals and end access.
Start with published archives and rights metadata, then extend to unpublished work only on private deployments where prompts, documents and embeddings stay inside a restricted workspace.
Keys, contributors and least privilege
Issue a distinct API key per desk, workflow and environment, and keep investigation workspaces separate from general newsroom retrieval. Store keys in a secrets manager, rotate on a schedule, and revoke them when freelancers, stringers or vendors finish an assignment. Enterprise SSO and RBAC options keep console access aligned with newsroom role changes.
Never place source names or unpublished excerpts in prompts where retrieval can supply only the passage a task requires.
Residency, retention and provenance
Decide where processing happens and how long records live. Region selection covers many needs; VPC, on-prem and air-gapped deployment covers investigations and archives with stricter handling requirements. Retention applies to prompts, outputs, logs and retrieval indexes, and rights metadata should travel with content so reuse stays traceable.
Log enough to reconstruct how a story or asset was produced: request ID, model and version, retrieved source identifiers, output and the reviewing editor. See AI audit logs for a schema.
Model governance and editorial review
Keep an approved model list with evaluation evidence and re-test when versions change. Ground background material in cited archives so journalists can verify quickly, and mark AI-assisted drafts clearly in the workflow. Editorial judgement stays with named journalists and editors, and the checklist should state what may be automated, such as tagging, and what may not, such as publication decisions. Audio and moderation endpoints are coming soon; plan transcription and comment pipelines around that status.
Honest comparison
| Control area | Plugsky capability | Common gap | Owner |
|---|---|---|---|
| Workspace separation | Separate keys and indexes per desk or investigation | One archive for everything | Editorial operations |
| Identity | Scoped keys, rotation, enterprise SSO and RBAC options | Shared newsroom credentials | IT security |
| Data boundary | Cloud, VPC, on-prem or air-gapped deployment | Source material sent to public tools | Standards editor |
| Provenance | Rights and tagging records via JSON mode | Metadata lost in pipelines | Rights and archives |
| Retention | Configurable logging of requests and outputs | Prompts kept indefinitely | Legal and records |
| Review | Citations and structured output for editors | AI drafts published unreviewed | Managing editor |
Frequently asked questions
Does using Plugsky make us compliant?
No. Compliance is your program. Plugsky provides deployable controls - scoped keys, separated workspaces, private deployment, logging - that you document and audit against your own editorial standards and legal obligations.
Can investigation material stay off shared systems?
Yes. Use separate keys and retrieval indexes, and keep sensitive work on private deployment where prompts, documents and embeddings stay inside a restricted environment.
What should we log?
Request IDs, model names and versions, retrieved sources, outputs and the reviewing editor, retained under your policy so published work can be reconstructed.
How do we protect sources?
Keep source-identifying material out of general indexes and prompts, restrict retrieval to the investigation workspace, and avoid public endpoints for that class of data.
Can we transcribe interviews today?
Audio endpoints are coming soon. Today, use chat and retrieval over existing transcripts, and plan audio pipelines around that status.
Is moderation available for comments?
Moderation endpoints are coming soon. Until then, retrieval and prompting can assist review, but a human remains responsible for publication decisions.
Where should a pilot start?
Pilot on published archives and rights metadata with the free plan, validate tagging quality and logging, then extend to unpublished work on private deployment.