Key facts
| API compatibility | OpenAI-compatible chat, embeddings and function calling |
| Use cases | Literature triage, structured extraction, methods summaries, draft writing |
| Context | Long-context models for full papers and multi-document synthesis |
| Retrieval | plugsky-embed plus RAG to ground answers in your own corpus |
| Data controls | Scoped API keys, audit logs, region selection and retention settings |
| Deployment | Plugsky cloud, your VPC, on-prem and air-gapped options |
| Pricing model | Flat monthly self-serve plans with unlimited fair-use usage |
| Free tier | plugsky-micro and plugsky-lite on the free plan, no card required |
TL;DR
- Triage large literature sets before a researcher reads anything in full.
- Extract fields into a structured table instead of re-reading each paper.
- Ground findings in retrieved passages and keep citations in the output.
- Use region selection and on-prem deployment for unpublished datasets.
- Freeze prompts and model versions so results stay reproducible.
How it works, step by step
- Define the extraction schema first: the fields, units and accepted values you need per paper.
- Run a triage pass to rank papers by relevance so deep review time goes to the right ones.
- Extract structured data with a strict schema and validate a sample by hand.
- Ground synthesis in retrieved passages with citations, and keep the passage ids in your notes.
- Pin the model version and store prompts with each results table for reproducibility.
- Move sensitive or embargoed corpora to a region or on-prem deployment before scaling up.
Try it yourself
Open the context window comparison →
Where AI helps the research workflow
Research is a read-extract-synthesise loop, and each stage has a narrow AI task:
- Literature triage: score abstracts and full texts for relevance to a specific question.
- Structured extraction: pull sample sizes, methods, outcomes and limitations into a review-ready table.
- Methods summarisation: condense a protocol or paper into a consistent short form.
- Draft support: produce an outline or first pass for the researcher to rewrite.
The researcher remains the author and the judge of quality.
Accuracy, citations and reproducibility
Language models can produce plausible but unsupported statements, so retrieval and citation checks matter more than fluency. Index your corpus with embeddings, require every claim to cite a retrieved passage, and include a verification step that flags claims with no supporting text. Record the model version, prompt and parameters alongside each output so a result can be reproduced later. For extraction, sample a percentage of records and check them by hand; any error rate above your threshold means the schema or prompt needs work, not more volume.
Handling sensitive and embargoed data
Unpublished results, human-subject data and grant material carry different constraints. Choose the processing region deliberately, set retention to the minimum that supports your workflow, and keep identifiers out of prompts. Where a funder, ethics board or collaborator restricts processing locations, move the workload to your own VPC, an on-prem install or an air-gapped environment. Because Plugsky is OpenAI-compatible, the pipeline you validated on the free plan still runs in those environments without a rewrite.
A practical starting point
Begin with one question and one corpus on plugsky-micro or plugsky-lite, then use the 14-day full-access trial to compare a frontier model and a long-context model on the same extraction task. Score field accuracy, missing-citation rate and researcher minutes per paper. Self-serve plans are flat monthly with fair-use usage, so a large screening run does not produce a variable bill. Keep evaluation sets as your corpus grows, and re-test whenever you change models.
Honest comparison
| Capability | Plugsky | Typical per-token API | Building in-house |
|---|---|---|---|
| API compatibility | OpenAI-compatible chat, embeddings and tools | Usually compatible | Full rewrite |
| Long context | Long-context models for full papers and synthesis | Varies by provider | You host and tune models |
| Deployment | Cloud, VPC, on-prem and air-gapped | Mostly cloud-only | You operate GPUs and serving |
| Pricing | Flat monthly self-serve, fair-use usage | Per-token, harder to budget | GPU plus operations cost |
| Model choice | 30+ models behind one API | Varies by provider | You host every model |
Frequently asked questions
Can we keep our existing OpenAI SDK code?
Yes. Plugsky exposes an OpenAI-compatible API, so existing pipelines change only the base URL and model name.
Is there a free plan?
Yes. The free plan includes plugsky-micro and plugsky-lite with no credit card, which is enough to prototype triage and extraction.
How do we keep citations honest?
Retrieve passages with embeddings, require each claim to cite a passage id, and run a verification pass that flags unsupported statements before a human reviews.
Can we use it with unpublished data?
Use region selection and minimum retention, keep identifiers out of prompts, and move to a VPC, on-prem or air-gapped deployment when a funder or ethics board restricts processing locations.
Which model should we use for long papers?
Long-context models handle full papers; for multi-document synthesis, combine retrieval with a strong model and compare on your own corpus before committing.
How does pricing work?
Self-serve plans are flat monthly with unlimited fair-use usage. See the live pricing page for current plans and enterprise options.
Which endpoints are live today?
Chat, streaming, JSON mode, function calling, embeddings, RAG and agents are live. Audio, images, moderation, files, batch, fine-tuning, assistants and responses are coming soon.