Closed alpha · University of Münster

The end-to-end SLR workspace that shows its work.

Search, screen, extract, synthesise and report. Every AI recommendation carries a confidence score, a per-criterion justification, and a link back to the source. The system recommends. You decide.

Free for academic useOpen coreEU-hosted (Münster)Self-hostable
Title / Abstract ScreeningPaper 5 of 432

LitFlow: An integrated, AI-augmented systematic literature review platform

Näscher, Strohmann & vom Brocke

2026 · DESRIST · Design Science Research in Information Systems and Technology

Abstract

Systematic literature reviews (SLRs) are central to rigorous research but remain resource-intensive and dependent on fragmented toolchains. This paper presents IN1LitFlow, a web-based platform for AI-augmented systematic literature reviews. Its augmentation approach provides IN2AI recommendations with confidence scores, justifications and source references, while final decisions remain with the researcher — developed following the echeloned design science research methodology.

Screening Assistantrecommends
Likely relevantMistral Medium 3.5
Criteria met95%
Confidence97%
Reasoning

Presents an AI-augmented SLR platform whose augmentation approach keeps final decisions with the researcher — a methodological contribution that satisfies the reflection criterion.

Criteria assessment6 / 6 met
INCL.Met

Addresses AI/LLMs/GenAI as a tool or method within qualitative or text-based research methods.

INCL.Met

Makes a methodological, conceptual or normative contribution to this topic.

EXCL.Does not apply

Uses AI merely as a tool to conduct its own substantive empirical study.

Screening decision
IncludeI
ExcludeE
MaybeM
Four things we do differently

An SLR tool that's honest about the AI inside it.

Existing tools cover parts of the review and have started adding AI — but configurability and workflow coverage tend to be inversely related, and AI outputs typically lack per-decision transparency. LitFlow refuses that trade-off.

01End-to-end

One workspace, five phases.

Search, screen, extract, synthesise and report — without leaving the document model. Every paper carries its full history, every phase shares the same audit trail.

SearchDatabases
ScreenAI + you
ExtractSchemas
SynthesiseThemes
ReportAudit trail
02Augmentation

AI assists. You author the decision.

For every record the AI proposes a per-criterion verdict with a confidence score, a justification, and a link to the source. The decision, the interpretation and the authorship stay yours.

Design principles96%
Methodological contribution91%
Source: §1, p.2 — "seven design principles"
03Audit

Provenance you can audit. Methods you can publish.

Every AI suggestion, every human decision — logged with timestamp, actor and rationale. Export a methods report that includes Cohen's κ between AI and reviewer.

14:02:17AIrecommend include · 97% · 6 of 6 criteria
14:02:43HNopened source · IN1 highlight
14:03:09HNconfirmed include · note: "EJIS, peer-reviewed"
14:03:09AIκ updated
04Open

Provider-agnostic, self-hostable, open core.

The LLM layer is abstracted. Swap commercial models for self-hosted ones — or vice versa — without touching application code.

Commercial model · activeactive
Azure / EU endpoint · configuredconfigured
Self-hosted model · localconfigured
Walk through it

Five phases. Click any to see what it looks like in the app.

Each phase carries its own AI surface — but they share the same document model, the same audit trail, and the same researcher-in-the-loop pattern. Pick a step.

The AI explains itself. You decide.

For each paper, every inclusion and exclusion criterion gets its own verdict — with a confidence score, a justification, and a link back to the passage that informed it.

  • Per-criterion verdicts — accept the whole or override individual lines
  • Confidence scoring against your project's reference set
  • Cohen's κ updates live as you and the AI work through the corpus
AI · proposal97% confidence
Design principles✓ matches · 96%
Proposes and evaluates seven design principles for GenAI creativity support tools.
Methodological contribution✓ matches · 91%
Makes a normative contribution to AI-augmented research methods.
Excludes: AI as bare tool✓ does not apply · 99%
Paper reflects on the method, not merely applies it.
Positioning

Why we built another one.

As of early 2026, no existing tool we evaluated combines transparent, configurable AI augmentation across the full SLR workflow — from search through reporting — with researcher authority and auditability preserved at every stage.

fullpartialnone
RayyanCovidenceASReviewEPPI-ReviewerLitFlow
OSI-approved open-source core
Self-hostable
AI / ML screening support
Per-criterion AI justification
End-to-end: search → report
EU-hosted instance
Free for academic use
  • EPPI-Reviewer source released Sept 2024 under FSL-1.1-MIT (source-available, not OSI-approved); self-hosting requires Microsoft SQL Server.
  • Covidence does not support literature identification / search — import-only from external databases.
  • Rayyan: US-incorporated, GDPR-compliant via SCCs and an EU representative, but no documented EU data-residency option.
  • EPPI-Reviewer is hosted at UCL (UK); the UK holds a GDPR adequacy decision but is outside the EU.
Built on research

Grounded in evidence.

LitFlow grew out of a research programme at the University of Münster. Its design draws on a requirements study with practising researchers, a formative think-aloud evaluation, and ongoing work published at peer-reviewed venues.

178
researchers in requirements study
5
formative evaluation participants
DESRIST 2026
peer-reviewed paper
Open core
license under review
Evaluation

What researchers told us.

From the formative think-aloud evaluation with five IS researchers covering the full workflow — from project creation through reporting.

"Honestly, the AI wasn't even the best part. The best part is finally having one tool for the whole review — instead of PDFs in one place, analysis in another, mapping somewhere else, and your synthesis scattered across spreadsheets."
P1
Anonymised participantIS researcher · think-aloud evaluation
"It's an assistant that takes the unpleasant work off your hands. You can do everything yourself, or lean on the AI when you want to — you always keep the upper hand."
P2
Anonymised participantIS researcher · think-aloud evaluation
"Surprisingly restrained. You don't really perceive it as an AI tool — and that's exactly what I want. If I feel like using the AI, I use it. If I don't, I can opt out entirely."
P3
Anonymised participantIS researcher · think-aloud evaluation · on the role of AI
Common questions

Things researchers ask first.

If something here isn't covered, the community Discussions are open — and the paper goes deeper still.

Is LitFlow free?

Yes — free for academic use. LitFlow is an academic research artifact with an open-core release planned. There is no commercial tier.

Which AI models does it use?

LitFlow is provider-agnostic. Bring your own key (BYOK) for commercial models — Gemini, Claude, OpenAI or Mistral — or use the open-source models hosted by the University of Münster. You can swap between them without touching application code.

Hosted instance or self-hosting?

The closed alpha runs on the instance hosted on University of Münster infrastructure in the EU. An open-core release is planned so institutions can later run it themselves; the exact license is still being finalised.

Does the AI make the decisions?

No. The AI proposes — a per-criterion verdict with a confidence score and a justification. Inclusion, exclusion and interpretation stay human decisions, always.

How is my data handled?

The hosted instance runs on EU infrastructure under GDPR. Personal data is pseudonymised in research outputs. See the privacy notice for the full detail.

Which disciplines is it for?

LitFlow is rooted in information systems research but the workflow is used across medicine, software engineering and the social sciences.

How do I take part?

LitFlow is in closed alpha. Read what the alpha involves, then register on the application — maintainers approve new cohort members.

Join the alpha. We'll make the methods report easy.

Free for academic use. The closed alpha runs on the University of Münster's hosted instance; an open-core release will follow.