About Compendium

Better tools won't fix bad methodology. Better methodology, built into the tool, might.

Compendium embeds gold-standard reporting guidelines and a systematic-review workflow modeled on the Cochrane Handbook into the software you use every day. It also runs AI on your own computer, so your data never has to leave it. Compendium is an independent tool, not affiliated with, endorsed by, or produced by Cochrane, PRISMA or JBI.

The problem

Research methodology isn't broken. It's just unreasonably hard to follow.

The standards exist, from PRISMA to the Cochrane Handbook. The problem is what happens between the standard and the finished manuscript:

  • Incomplete PRISMA flow diagrams, protocols that omit pre-specified outcomes, risk of bias assessed holistically instead of domain by domain, meta-analyses that pool clinically incompatible studies because nothing stopped it.
  • Tooling makes it worse: a typical review touches a reference manager, a screening tool, a spreadsheet, a statistics package, a diagram generator and a word processor. Each handoff is where numbers drift and steps get skipped.
  • Clinical research often involves sensitive, ethics-governed or unpublished data that doesn't belong on a third party's server, yet most tools are cloud-only.

Compendium exists to solve the three together.

What we believe

What we build into the software, and what we leave to you.

I

Rigour by default, not by luck

Every workflow is designed with reference to the Cochrane Handbook or the relevant gold-standard reporting guideline. Required fields are visible, not hidden, so you can't silently skip eligibility criteria or exclusion reasons.

II

AI assists. You decide.

Every AI suggestion shows a confidence indicator and an explanation. Nothing applies without your confirmation: judgment in evidence synthesis belongs to the review authors, not to automation.

III

Teach, not just check

A checklist that flags "item not reported" helps you pass review. Explaining what it means and why helps you become a better researcher. Guidance is written with reference to the Cochrane Handbook, PRISMA 2020 E&E and the JBI Manual.

IV

Privacy by architecture, not by policy

AI models run on your own computer; your research data stays in a local database by default. The data doesn't leave because the software doesn't send it, not because of a policy. There's no cloud AI option today: every model runs on your machine.

V

Always exportable, no lock-in

At any stage, protocol, screening, extracted data, results, PRISMA diagram, you can export to standard formats (RIS, CSV, PDF, DOCX). Your data and decisions are yours.

AI & scientific integrity

A careful note on AI in research.

AI in Compendium suggests screening decisions from your eligibility criteria, pre-fills extraction fields from article text, and drafts prose from data you've already validated. It proposes; you decide. Rigour comes from the methodology, not the AI. Speed is useful. Speed is not rigour.

  • Risks include fabricated references, hallucinated data and confident-sounding text that's wrong.
  • Mitigation: every suggestion shows a confidence score and a justification you can check against the source; AI-assisted content is logged and flagged so you can disclose AI use, as journals increasingly require.
  • Local models are smaller than the largest cloud models. We accept that trade-off because your data staying on your machine matters more, and the final call, and the accountability for it, is always yours.
Who's building this

Built by a researcher, for researchers.

Compendium is an early-stage, independent project, not a venture-backed team of fifty. One researcher hit the same methodological and tooling problems you're hitting, and built the tool that should exist, shaped by feedback from the researchers who use it.

Vitor Pio Daldegan is a medical student at the University of Campinas (UNICAMP) and a researcher in clinical evidence synthesis, working on meta-analysis, Bayesian sensitivity analysis and E-values applied to perinatal psychiatry and neurodevelopment. An abstract from this work was accepted for the World Congress of Psychiatry (WCP). He founded Compendium to bring local AI into the clinical researcher's workflow.

The product
A working Windows MVP

A functional desktop app (.exe), covered by 48 automated tests. Early access, not a finished product.

In the field
Early access

No paying customers yet. The core workflow is real and usable today.

How to start
No waitlist

Write to askcompendium@gmail.com. No forms, no queue, a person replies.

Today's scope: the current product is focused on systematic reviews (PRISMA), end to end, covering protocol, search, AI-assisted screening, extraction, risk of bias, meta-analysis and PRISMA reporting. More study types are on the roadmap.

Care about methodology? Help shape the tool.

Compendium is in early development. The first cohort of researchers helps decide what comes next.

Free for early users. No credit card.