VerdictPal
Join
← All tools
Cite
DOSSIER · Systematic review · LIGHTWEIGHT · VERIFIED 12 Oct 2026
ASReview LAB70LightweightLocal active-learning screening for systematic reviewsVerified 2026-10-12

Dossier · Systematic review

ASReview LAB

The model orders the reading pile. It does not certify that you are done.

What is ASReview LAB?

Decision

Local active-learning screening for systematic reviews

Choose it when
  • PhD students or librarians with thousands of titles and abstracts to screen against defined inclusion criteria
  • Reviews that need inspectable methods and records kept on a local machine or an institution's own server
Skip it when
  • You need full-text review, extraction, risk-of-bias assessment and PRISMA reporting in one platform
  • You cannot install Python or Docker and have no institutional host
Cost
Free, open source under Apache-2.0.
Privacy
ASReview LAB runs locally, or on a server you or your institution host.

Watch for A run of irrelevant records does not prove there are no relevant studies left. Stopping early can miss studies; the dashboard has no default threshold and the docs say the ideal threshold is still being researched.

Compare with RayyanHow we score

Synthesized8 sourcespricing checked 12 Oct 2026privacy checked 12 Oct 2026

What these labels mean
Editorial fit
0–100 desk score for research-workflow fit. Not stars, not a benchmark.
Evidence level
How close the desk got: self-run, hands-on, synthesized, or vendor claim.
Dossier depth
Flagship, Solid, Lightweight, and Draft describe how finished the page is — not which tool is best.
Verification date
When pricing or privacy was last opened at the source. Not a re-score of the whole dossier.
Benchmark status
Pending means no relevant first-party test is published yet. It is not a negative score.

Open the glossary · Read the methodology

ASReview is free, open-source software for AI-aided screening in systematic reviews, released under the Apache-2.0 licence and coordinated at Utrecht University. Its models prioritise records using your relevance labels; a human makes the screening decisions. The repository describes monitoring progress and deciding when to stop. Checked 13 October 2026.

Role: Local active-learning screening for systematic reviewsCategory: Systematic reviewSynthesized
ASReview LAB at a glance, with the date each field was checked.
FieldValue
Editorial fit70 out of 100
Dossier statusLightweight
SetSystematic review
Role in a workflowLocal active-learning screening for systematic reviews
PricingASReview LAB Free
Training on your dataIn the standard local workflow, ASReview trains a model on your project's relevance labels; this is local active learning, not vendor-cloud training. The developers state that ASReview LAB collects no usage or user data. On a self-hosted server, your operator controls the data environment. Optional extensions and hosted demos need separate checks; their external API use is not verified here. Sources: ASReview introduction and PyPI, checked 12 Oct 2026 (linked in Sources).
Last verified2026-10-12

Public facts · vendor & repo

List prices and pay-as-you-go entry points we can cite without running our own bench. Each tile links to a source when possible.

Free, Apache-2.0Software licencePyPI, 12 Oct 2026 ↗

Limits & product surface

Non-price vendor claims — multipliers, caps, and API scope. Detailed matrices live in subscription and SDK sections below.

Version and runtime
3.0.8; Python 3.10 or later.PyPI ↗
Source code
Open source at github.com/asreview/asreview, linked by PyPI.PyPI project links ↗
Screening decision
The model ranks; the reviewer labels relevance.Current documentation ↗
Stopping threshold
Not set by default; the ideal threshold is still under research.Progress documentation ↗
AI training
The local model learns from project labels; developers state no usage or user-data collection. Extension API use is unverified.PyPI and introduction ↗
RIS export
Only available if the imported dataset was RIS.Export documentation ↗

How it works · agent loop

The public positioning for this product — the loop we score against on VerdictPal.

Prepare the search set

Define inclusion criteria and collect records outside ASReview. Check titles, abstracts and duplicates before importing CSV, RIS or Excel; the model cannot repair a search that missed a study.

Install and initialise

PyPI requires Python 3.10+ and gives pip install asreview as the install command. Start the local interface with asreview lab. Create a project and prepare relevant and irrelevant training labels; SAFE recommends a random warm-up sample rather than only familiar papers.

Screen and check the stopping plan

Apply your criteria to each proposed record while the model updates the ranking. Agree a stopping strategy, monitor the dashboard and keep the decision human. A long irrelevant streak is a signal, not proof of complete recall.

Export decisions, not a completeness claim

Export labels and the project, retain the dataset and settings, and document the stopping rationale. The export distinguishes unseen records from irrelevant ones. Move eligible records to full-text review and report the AI-assisted process transparently.

Who it fits, where it fails

Best for

  • PhD students or librarians with thousands of titles and abstracts to screen against defined inclusion criteria
  • Reviews that need inspectable methods and records kept on a local machine or an institution's own server
  • Students with no licence budget who can install Python or arrange a self-hosted setup
  • Methods work comparing active-learning models on a fully labelled dataset

Avoid if

  • You need full-text review, extraction, risk-of-bias assessment and PRISMA reporting in one platform
  • You cannot install Python or Docker and have no institutional host
  • Nobody on the review team can justify and document a stopping rule
  • You expect the tool to find papers that your database search never retrieved

Strengths

  • Title and abstract screening with human relevance labels
  • Active-learning ranking updated from your labels
  • CSV, RIS and Excel dataset import
  • Prior-knowledge labels to initialise the model
  • Preconfigured ELAS models and custom model components
  • Simulation on fully labelled datasets
  • Editable labels, tags and screening notes
  • Automatic hiding of duplicate titles and texts in version 3
  • Progress dashboard with a reviewer-set stopping threshold
  • Export of labels and records to CSV, TSV, Excel or RIS
  • Export and re-import of an entire .asreview project
  • Self-hosted LAB Server with authentication and multiple reviewers

Weaknesses

  • A run of irrelevant records does not prove there are no relevant studies left. Stopping early can miss studies; the dashboard has no default threshold and the docs say the ideal threshold is still being researched.
  • The model can only rank the records you import. An incomplete search or missing titles and abstracts can leave relevant studies outside the useful screening pool.
  • Your labels train the ranking. The SAFE paper warns that a seed set chosen only from familiar key papers can bias the model, and recommends a random training sample.
  • Free software still needs installation and maintenance. The current package requires Python 3.10 or later; a shared server also needs someone responsible for hosting and access.
  • Screening is not the whole review. Full-text eligibility checks, extraction, risk-of-bias assessment and reporting still need a separate process.
  • Automatic duplicate hiding in version 3 changes what appears during screening. PyPI says hidden duplicates can be included in exports; do not treat the visible screening count as a complete deduplication report.

How it could improve

  • Metadata quality could be improved with a verification layer that cross-checks imported records against authoritative databases before they enter the library.
  • Network quality depends on starting-point relevance. ASReview LAB could improve by offering a "seed quality" indicator, how well-connected the starting paper is, and suggesting better-connected alternatives when the seed is peripheral.
  • Extracted values that require manual verification suggest the extraction pipeline needs a confidence-checking layer, cross-reference extracted claims against the original passage before presenting them as fact.
  • Extracted values that require manual verification suggest the extraction pipeline needs a confidence-checking layer, cross-reference extracted claims against the original passage before presenting them as fact.
  • Duplicate and preprint records inflate counts and confuse search. A deduplication pass that clusters preprint–published pairs would clean the corpus significantly.
  • Setup friction scores 40/100, the weakest dimension on this dossier. Addressing this would meaningfully raise the editorial fit.
  • Reliability scores 0/100, the weakest dimension on this dossier. Addressing this would meaningfully raise the editorial fit.

Search modes · 4 lenses

One input box, many retrieval postures. Filter by tier.

01Free

Active-learning screening

The model reorders records from your relevance labels; open source, runs locally.

02Free

Duplicates & tags

v3 hides duplicates automatically and supports editable tags.

03Free

Stopping decision

You decide when to stop; the ranking does not certify completeness.

04Free

LAB Server

Self-hosted multi-user setup; the operator owns storage and access.

Competitive lens

Where ASReview LAB wins for cited research — and where a rival still belongs in the stack.

ASReview's documented distinction is free, inspectable active learning on your machine or your own server. It is not necessarily solo-only: LAB Server supports multiple reviewers. This dossier does not establish blinded-review or conflict-resolution parity with Rayyan.

ASReview LAB

  • You need free software, local records and inspectable model components.

Rayyan

  • Use Rayyan as a comparison candidate if ready-to-use team screening matters more; verify its current collaboration features and plan limits separately.

Scores and evidence

Metric lab · 16 dimensions

Click a tile for the editorial note. Color follows score: coral, yellow, mint.

Shape

Avg 70 · 0–85

70Editorial fit

Provisional, source-synthesized; no hands-on testing. Weighted mean of 13 dimensions using the published rubric, including the flagged reliability zero; based on PyPI, documentation and the methods/SAFE papers, not a screening trial.

By group

Fit75
  • Editorial fit70
  • Wedge task fit80
  • Feature depth75
Cost83
  • Free-tier utility85
  • Cost-to-value85
  • Opportunity cost80
Trust78
  • Source grounding75
  • Privacy posture85
  • Failure transparency80
  • Evidence strength70
  • Transparency80
Workflow50
  • Integration reach65
  • Setup friction40
  • Reliability0
  • Competitive position65
  • Data portability80

Editorial evidence · 3 entries

Task

Check who makes screening decisions

Current documentation assigns ranking to the model and relevance labels to the human. Simulation uses a dataset whose labels are already known.

↳ Documentation synthesis, not a screening trial or a measured recall result.

Task

Check the stopping threshold

The progress docs confirm no default threshold. The SAFE paper proposes four phases and explicitly warns that its numeric examples are not exact rules.

↳ No stopping rule was tested on a VerdictPal dataset; no recall or time-saving percentage is claimed.

Task

Check version, licence and data handling

PyPI lists 3.0.8, Apache-2.0 and Python 3.10+. PyPI and the docs state no usage or user-data collection; the docs describe LAB Server as self-hosted.

↳ No installation, network capture or extension audit was run.

Test scenarios · hands-on lab

How we exercised the product. Step through each run before you trust the scores.

Test run

Step 1 of 4

Write inclusion criteria and check the imported title/abstract set.

Research log · desk notes

What we learned while building this dossier — not vendor copy.

PyPI ↗

Use the current package, not an old install requirement

The brief referenced version 2.0. Live PyPI lists 3.0.8, released 18 Jun 2026, with Python 3.10+. Its v3 summary documents automatic duplicate hiding and editable tags; no broader changelog claims are made here.

Systematic Reviews, 2024 ↗

SAFE is a procedure, not a magic cutoff

The full 2024 paper was read. It proposes warm-up screening, active learning, model switching and an exclusion quality check. It calls its numbers examples, not exact rules, and says no direct comparison with existing stopping heuristics was presented.

Nature Machine Intelligence, 2021 ↗

Old mode names are not the current menu

The 2021 methods paper describes oracle, simulation and exploration modes and deliberately excludes author and citation-network features to avoid authority bias. Current 3.0.8 documentation presents Review and Simulate. The historical paper is not evidence that every current extension shares the same feature choices.

What it costs, what it keeps

Pricing deck · checked 2026-10-12

ASReview LAB

$0

Free, Apache-2.0

  • Local Python installation; Docker is also documented.
  • No paid software tiers found; hosting and support are your responsibility.
  • Current PyPI version: 3.0.8, Python 3.10 or later.

RAW · checked 2026-10-12: Free, open source under Apache-2.0. No paid software tiers found in the checked sources. PyPI lists ASReview LAB 3.0.8, released 18 Jun 2026, requiring Python 3.10 or later. Local installation has no licence fee; your own hosting, IT support and setup time are separate costs.

Privacy deep-dive · checked 2026-10-12

ASReview LAB runs locally, or on a server you or your institution host. Its active-learning model trains on your project's relevance labels. The developers state on PyPI and in the docs that the software collects no usage or user data; the 2021 methods paper describes records and labels staying on the user's computer without third-party processing. This describes the standard local workflow, not a network audit. LAB Server is self-hosted: its operator, not an assumed vendor cloud, controls hosting and access. Optional extensions and hosted demos need their own data-flow checks; external API use has not been verified here. The homepage's GDPR and AI-regulation claims are developer statements, not a certification verified by VerdictPal.

Training on your dataIn the standard local workflow, ASReview trains a model on your project's relevance labels; this is local active learning, not vendor-cloud training. The developers state that ASReview LAB collects no usage or user data. On a self-hosted server, your operator controls the data environment. Optional extensions and hosted demos need separate checks; their external API use is not verified here. Sources: ASReview introduction and PyPI, checked 12 Oct 2026 (linked in Sources).
EU data residencyUnknown
SOC 2 attestationUnknown
Local-first by defaultASReview LAB runs locally; LAB Server is self-hosted by you or your institution. The local-data statement covers that setup, not an unverified hosted demo or every optional extension.

Under the hood

Vendors named on the product about page — useful for procurement and privacy reviews.

Record prioritisation and labels
ASReview LAB
Shared, institution-hosted screening
ASReview LAB Server

Alternatives and context

Workflow roles

How this tool fits into a composed research stack:

Prioritise imported titles and abstractsLabel relevance against review criteriaDocument a stopping decisionExport screening decisions for the next review stage

Deep panes · 9 lenses

Editorial lenses only. Subscription and API pricing live in their own sections above.

What is ASReview LAB?

ASReview LAB is Utrecht-coordinated, open-source software for screening a set of records you already collected. You mark titles and abstracts relevant or irrelevant; active learning changes the reading order. It does not write a literature review or replace the database search. Scores are provisional and come from documented sources, not hands-on testing.

  • PhD students or librarians with thousands of titles and abstracts to screen against defined inclusion criteria
  • Reviews that need inspectable methods and records kept on a local machine or an institution's own server
  • Students with no licence budget who can install Python or arrange a self-hosted setup
  • Methods work comparing active-learning models on a fully labelled dataset

Sources and provenance

Source ledger · 8 checked

Quality gate · Lightweight

Evidence ready

Verdict history

  • Moved out of draft to lightweight with provisional scores from documented sources, not hands-on testing. Reliability remains unrated; no recall or time-saving result is implied.
  • Added a bilingual draft covering local active learning, free Apache-2.0 software, current 3.0.8 requirements and the reviewer’s stopping decision. Desk scores remain open.
SynthesizedLightweight
Visit ASReview LABdirect link

Related dossiers

Systematic review
Rayyan
60/100
ROLESystematic review screening and collaborative triage

Prosumer SaaS · Text · Data

Cloud systematic-review screening workspace for title/abstract triage, deduplication, labels, PICO extraction, PRISMA flow support, collaborators, mobile screening, and AI-assisted inclusion decisions.

Watch outYou need collaborators to inherit premium features on reviews you do not own without an Institutional plan

SynthesizedSolid
Screening titles abstractsTeams that need
From$4.99/MOper seat
Editorial fit
60
Source quality
55
Citation honesty
66
Privacy controls
53
Value for money
78
Data portability
49
Research assistant
Elicit
59/100
ROLEPaper discovery, evidence-table extraction, and systematic-review support

Prosumer SaaS · Text · Data

AI research assistant for searching, summarizing, chatting with papers, screening studies, extracting fields, generating research reports, and supporting systematic reviews across large scientific corpora and uploaded documents.

Watch outYou need a cheap casual search tool rather than a high-cost research workflow

SynthesizedFlagship
Building structured evidenceEarly-stage systematic review
From$132/YRper user
Editorial fit
59
Source quality
52
Citation honesty
70
Privacy controls
46
Value for money
56
Data portability
55

Browse the atlas

Questions this dossier answers

Every answer below is assembled from the dated fields on this page. Nothing is written separately for search.

Is ASReview free?

ASReview is free, open-source software for AI-aided screening in systematic reviews, released under the Apache-2.0 licence and coordinated at Utrecht University. Its models prioritise records using your relevance labels; a human makes the screening decisions. The repository describes monitoring progress and deciding when to stop. Checked 13 October 2026.

Who makes ASReview?

ASReview's website says the project is coordinated at Utrecht University in the Netherlands.

What licence does ASReview use?

ASReview's GitHub repository lists the Apache-2.0 licence.

Does ASReview send my data to the cloud?

ASReview's website says you retain control of your data, and its repository says ASReview LAB collects no usage or user data. That does not verify the data flows of every hosted installation or optional extension.

Can ASReview screen papers on its own?

The ASReview repository describes models that learn from your relevance labels and prioritise records. You label records as relevant or not relevant, monitor progress and decide when to stop.

Is ASReview LAB worth using?

ASReview LAB scores 70 out of 100 on editorial fit and carries Lightweight dossier status. Its job in a research workflow is: Local active-learning screening for systematic reviews.

Who is ASReview LAB best for?

ASReview LAB earns its place when you need:

  • PhD students or librarians with thousands of titles and abstracts to screen against defined inclusion criteria
  • Reviews that need inspectable methods and records kept on a local machine or an institution's own server
  • Students with no licence budget who can install Python or arrange a self-hosted setup
  • Methods work comparing active-learning models on a fully labelled dataset

When should you not use ASReview LAB?

Skip ASReview LAB in these cases:

  • You need full-text review, extraction, risk-of-bias assessment and PRISMA reporting in one platform
  • You cannot install Python or Docker and have no institutional host
  • Nobody on the review team can justify and document a stopping rule
  • You expect the tool to find papers that your database search never retrieved

What goes wrong with ASReview LAB?

The dossier publishes 6 failure modes for ASReview LAB, and they stay published whether or not the vendor likes them:

  • A run of irrelevant records does not prove there are no relevant studies left. Stopping early can miss studies; the dashboard has no default threshold and the docs say the ideal threshold is still being researched.
  • The model can only rank the records you import. An incomplete search or missing titles and abstracts can leave relevant studies outside the useful screening pool.
  • Your labels train the ranking. The SAFE paper warns that a seed set chosen only from familiar key papers can bias the model, and recommends a random training sample.
  • Free software still needs installation and maintenance. The current package requires Python 3.10 or later; a shared server also needs someone responsible for hosting and access.
  • Screening is not the whole review. Full-text eligibility checks, extraction, risk-of-bias assessment and reporting still need a separate process.
  • Automatic duplicate hiding in version 3 changes what appears during screening. PyPI says hidden duplicates can be included in exports; do not treat the visible screening count as a complete deduplication report.

What are the alternatives to ASReview LAB?

The closest options to ASReview LAB are Rayyan, Covidence, Elicit, EPPI-Reviewer. Each one that has a tool in the atlas is linked from this dossier, with a head-to-head comparison.

The Pack · Editorial newsletter

Know when the rules move.

One short email when a university changes its AI rule for theses, or a research tool changes its pricing or privacy terms. No tracking, no third-party analytics. Unsubscribe in one click.

The thesis AI checklist is free to use, no signup needed →