SysRevAI, step by step

Getting started

SysRevAI is a platform installed on your institution’s own server for systematic and scoping reviews with AI assistance. This chapter takes you from zero to having the software downloaded and the licence in hand.

What SysRevAI is

It covers the whole review workflow: protocol, reference import and deduplication, title/abstract screening, full text, data extraction, risk of bias, GRADE synthesis and exports (PRISMA 2020 diagram, Word, Excel, RIS). AI is there as an assistant: it prioritises, suggests and summarises, but every decision is made by a person and logged.

Data never leaves your server. AI calls use your own key and can be switched off review by review.

Try it without installing

To see how it works before installing anything, create a 30-day trial account at sysrevai.com/prova. You can invite up to 5 people and run a real review (up to 3 reviews, 500 MB of PDFs and a small AI budget per team). On day 20 you get a notice; on day 30 the account is blocked and you can only export the data, and 7 days later it is deleted. Before that, export everything from Profile → My data and import it into your own installation.

What you need

The licence number

Its format is SRAI-XXXXX-XXXXX-XXXXX-XXXXX. It is free, identifies your installation for updates and is valid for one domain and up to 10 active users. The installer asks for it at step 7 and it is mandatory: it is always validated with sysrevai.com and cannot be skipped.

It works without an AI key too

You need no AI key to install or to run a review. The assisted features (screening and conflict proposals, prioritisation, pre-extraction, summaries, chat with the article) switch on once you add a Claude API key under Admin → Claude (AI). See chapter 3.

Link to this chapter →

Installing on the server

The web wizard does all the work in nine screens. No command line needed: upload the files, create the database and follow the steps.

Uploading the files

1. Unzip the package. Inside there is a single folder, sysrevai/.
2. Upload it whole to the server (SFTP, the hosting file manager or scp).
3. Point the site’s document root at sysrevai/public. In CloudPanel, Plesk or cPanel it is a site option; in Nginx or Apache, one line in the vhost.
4. Make sure PHP can write to storage/ and public/uploads/.

The wizard, step by step

After installing

• The cron job is essential. Add a cron that runs php /path/sysrevai/bin/worker.php every minute as the web site user. Without it there are no automatic backups, article retrieval and download, scheduled searches, long AI jobs or update checks. The installer gives you the exact line at the end and, while it is not running, administrators see an amber bar on every page with a “How to set up the cron job” button that leads to Admin → Maintenance → Scheduled tasks (cron): the line to copy, the steps over SSH or with a control panel (CloudPanel, Plesk, cPanel) and the status. Once the cron runs, the card says “Running” and the bar goes away.
• Sealed installer. Once finished, /install/ no longer opens. You may delete public/install/ if you prefer.
• Backups. Once the cron runs, the platform makes a full backup every 2 hours by itself (chapter 15).

* * * * * php /path/sysrevai/bin/worker.php >> /path/sysrevai/storage/logs/worker.log 2>&1

Common problems

Link to this chapter →

Initial configuration (Admin)

Once in, the admin panel (top menu → Settings) centralises everything that affects the installation: AI, e-mail, users, security and the defaults for reviews.

Recommended order

All sections

Platform roles

• Owner: everything. There is always at least one and they cannot delete themselves.
• Admin: settings and users, but cannot demote the owner.
• User: creates and takes part in reviews. Permissions inside each review are set by its coordinator (chapter 11).

Legal pages

Admin → Legal pages generates the privacy policy and terms of use from templates with the institution name, contact e-mail and date. Review them before opening registration: users accept them when creating an account.

Link to this chapter →

Creating a review and the protocol

A review is a team’s workspace: the protocol, the references, the decisions and the exports. Everything starts by defining the question.

Creating the review

1. On the dashboard, press New review.
2. Enter the title, the type (systematic or scoping) and the question framework: PICO for interventions or PCC (population, concept, context) for scoping reviews.
3. Choose the screening mode: single reviewer, double-blind, double-blind with third reviewer, or pilot. It can be changed later as long as there are no decisions.
4. Save. You can now invite the team (chapter 11) or go on with the protocol.

The protocol

The Protocol tab holds the question, inclusion and exclusion criteria, the exclusion reasons used during screening, the per-database search strategy and methodological notes. The criteria feed the keyword highlighting during screening.

Exclusion reasons

Define them before screening starts: every full-text exclusion needs one, and they are what appears in the PRISMA diagram. Defaults can be changed under Admin → Reviews.

The review dashboard

Each review’s home page shows the status of every phase (references, screening, full text, extraction, risk of bias, exports) with counters, who has work pending and recent activity. The phase bar at the top is the path: each step lights up when done.

Link to this chapter →

Importing references and duplicates

References come in by file, by direct search or by pasting text. Deduplication then leaves a single copy of each study.

Accepted formats

Import step by step

1. Inside the review, go to References → Import.
2. Pick the format, upload the file (or paste the text) and set the source (the database it comes from). The source is what shows in the PRISMA diagram.
3. Check the preview: each reference has a checkbox and a green mark when it has a DOI or PMID. Untick what you do not want.
4. Confirm. The references land in the title/abstract screening phase.

Direct search

From Search (top menu) you can query PubMed, CrossRef, OpenAlex and Europe PMC at once, see each result’s relevance and send them to the review without any file. Searches can be saved and re-run (chapter 14).

Deduplication

References → Duplicates looks for matches in three passes: exact identifiers (DOI, PMID), normalised title and similarity (Jaro-Winkler with authors and year). Each proposed pair is confirmed or dismissed; the reference with more metadata is kept and the other is marked as a duplicate, out of screening but counted in PRISMA.

Tags and local search

In the reference list you can filter by phase, decision, source, tag or text, and apply bulk actions (tag, move phase, export). Tags are free text and serve any classification of your own: "doubtful", "ask the library", "sub-study A".

Link to this chapter →

Title and abstract screening

The phase where the team decides, reference by reference, what goes on to full text. Two reviewers work blinded, a third settles disagreements and the AI can suggest a reading order.

The screening screen

On the left, the reference (title, authors, year, journal, abstract) with the protocol keywords highlighted. On the right, the inclusion and exclusion criteria. Below, the decision buttons. References come in import order or, if the coordinator enabled it, in the order the AI proposes.

Making a decision

1. Read the title and abstract.
2. Press Include, Exclude or Maybe. Keyboard: I, E, M.
3. If you exclude, pick the reason (optional in this phase; required at full text).
4. Move on with →. You can go back and change a decision while the phase is open.

Pending list and marks

• Pending list: the button next to Previous and Next opens the references you still have to screen, in queue order, with a filter by title, author or year. Click one to open it directly.
• Mark: the button next to the title marks the reference with a reason (Waiting for the PDF, Ask the team, Review later or Other) and a note. The whole team sees it, and it changes no decision and not the queue order. «Edit» changes it and «Remove mark» deletes it.
• Marked: in the Pending list, the «Marked» filter shows only your marked pending references, with the reason and the note. The number on the button tells you how many you have.

Modes and blinding

While the phase is open, no reviewer sees the others’ decisions. The coordinator does: the coordinator view shows everyone’s progress, pending references, and lets them share the work out.

Conflicts and the third reviewer

When two reviewers disagree, the reference appears under Conflicts. The resolver sees both decisions with their reasons and comments, decides, and can leave a resolution note. “Maybe” against “Include” is not a conflict: it moves on to full text. With the AI proposal button, the AI checks the article against each inclusion and exclusion criterion of the protocol and, inside the same card, recommends including or excluding with a confidence percentage, the reason and why it thinks the reviewers disagreed. “Copy to note” moves it into the resolution note, and the activity log keeps the proposal next to the final decision.

AI suggestions

While screening, the Suggest with AI button opens the same analysis as conflict resolution in the assessment column: the AI checks the article against the protocol and proposes include, exclude or maybe. The coordinator can also ask for a prioritisation of the whole queue (the most likely references come first) and an automatic screening of very low-scoring ones, always with a confidence threshold and reviewable. Everything goes into the AI declaration (chapter 13).

• Recommendation: include, exclude or maybe.
• Confidence as a percentage, with a green (high, from 80 %), yellow (medium, from 60 %) or red (low) bar.
• Reason in two to four sentences.
• Protocol criteria one by one: met, not met or not reported, with a short note.
• “Copy to note” and “Generate again” buttons. You always make the decision.

Inter-rater agreement

The Team panel shows Cohen’s kappa (two reviewers) or Fleiss’ kappa (more than two) computed over the references everyone has screened, and how many each person has left.

Link to this chapter →

Full text

Included references are now read in full. SysRevAI tries to fetch the PDF automatically and lets you decide next to the document.

Getting the PDFs

The automatic retrieval module queries, in order, legal open-access sources: PubMed Central, Europe PMC, Unpaywall, OpenAlex, Semantic Scholar, bioRxiv/medRxiv, arXiv, CrossRef and DOAJ. For each source it tries every open copy it knows of (the publisher’s and the repositories’), and when a link leads to a web page instead of the PDF it follows that page’s link to the document. It only uses official APIs and never bypasses a paywall or an “I’m not a robot” check. The cron worker processes the queue in the background. PubMed Central PDFs are taken from PMC’s open-data repository (NLM’s official channel for automated downloads), because the «PDF» links on PMC article pages ask for a browser check.

• Search full article: on the full-text screening screen, when an article has no text yet, the button searches every open source on the spot. If it finds the text but no PDF, it shows the text in a reader and keeps looking for a PDF. If it finds the article but the source does not allow automatic downloads, it says so (“We found the article at…, but it has to be downloaded by hand”) with the link to open it. If it finds nothing anywhere, it says so, gives you the publisher link (DOI) and asks you to upload it.
• PubMed Central articles restricted by the publisher: some articles can be read for free on PMC but the publisher lets no program download their text or PDF. In those cases the platform explains it: open the link, click “PDF”, save the file and upload it. After uploading you stay on the same reference.
• Queue: Full text → Queue shows what was found, what is missing and why.
• Manual upload: for articles without open access, upload the PDF you hold under your institutional licence. One by one or in bulk (a ZIP with the DOI or PMID in the file name).
• Coverage: Admin → Reports shows the share of references with full text per review.

Screening next to the document

The screen shows the PDF on the left (or the extracted text if the PDF could not be opened) and the criteria and buttons on the right. Here exclusion always asks for a reason: that is what PRISMA counts.

No PDF

If there is no way to get the document, mark the reference as "not retrievable". PRISMA counts it as a report not retrieved, apart from exclusions by reason.

Link to this chapter →

Data extraction

From the included studies the team extracts data with a template of the review’s own. It can be done in duplicate and compared.

Designing the template

1. Extraction → Template. Add the fields you need: text, number, date, option list, yes/no or long text.
2. Group them in sections (study characteristics, participants, intervention, outcomes).
3. Mark the required ones. Save. The template can be extended later without losing data.

Extracting

Each included study has a form built from the template. The PDF (or abstract) is shown alongside so you need not switch windows. Save often: the form remembers changes.

AI pre-extraction

With an AI key, the Fill with AI button proposes a value for each empty field from the full text, with the passage it came from. You verify, correct and save. Fields that already have a value are left alone.

Double extraction

If two reviewers extract the same study, the Compare view puts both forms side by side and marks the differences. The coordinator approves the final version, which is the one exported.

Exporting the data

From Export: CSV or Excel with one row per study and one column per field, and a long-format CSV (study × field × reviewer) ready for metafor, Stata or RevMan Web.

Link to this chapter →

Risk of bias

Each included study is assessed with the tool that fits its design. Judgements per domain are justified and exported for the report table.

Available tools

Assessing a study

1. Risk of bias → pick the study and the tool.
2. For each domain, answer the signalling questions, choose the judgement (low, some concerns, high) and write the justification quoting the text.
3. The overall judgement follows the tool’s algorithm; you can override it with a reason.
4. Save. A second reviewer can do their own assessment and compare.

Export

Export → Risk of bias produces a CSV with one row per study, reviewer and tool, and one column per domain with judgement and justification. Use it for the traffic-light plot with robvis or directly in the report.

Link to this chapter →

Synthesis, GRADE and exports

The work ends in documents: the PRISMA 2020 diagram, the summary-of-findings table with GRADE, the study tables and the bibliography of included studies.

PRISMA 2020 diagram

It is generated from the review’s counters: records per source, duplicates removed, screened, excluded, reports not retrieved, full-text exclusions by reason and included studies. Download it as editable SVG; any cell can be pinned by hand if an external count needs adjusting.

GRADE and summary of findings

1. Export → GRADE. Define the review’s outcomes.
2. For each outcome, rate the five domains (risk of bias, inconsistency, indirectness, imprecision, publication bias) and, where applicable, the factors that raise certainty.
3. The final certainty (high, moderate, low, very low) is computed and can be justified.
4. Export the summary-of-findings table as CSV or Word.

Every export

Public page

From the review you can create a read-only public link with the protocol and the list of included studies, handy for the article appendix or external collaborators. It can be revoked at any time.

Link to this chapter →

Team, roles and notifications

Each review has its own team with its own roles. Invitations arrive by e-mail or by link, and the platform tells you about everything that needs your attention.

Roles within a review

Inviting

1. Review → Team → Invite.
2. Enter the e-mail and the role. If the person already has an account they join the review directly; otherwise they get a link to create one.
3. Without SMTP, create an invitation link and send it through any channel. It expires in 7 days.

Sharing the work out

In the screening coordinator view you can assign references to specific reviewers (in blocks, equal shares or by hand) and follow each person’s progress. Unassigned references can be screened by any reviewer.

Notifications

The bell in the top menu flags new conflicts, mentions in comments, invitations and assigned tasks. Under Profile → Notifications you choose which ones to receive by e-mail and in the mobile app as well (chapter 18).

Comments and mentions

Any reference accepts comments. Write @name to alert a colleague. Comments show in screening, in conflict resolution and in the list.

Link to this chapter →

Tools: global search, citations, article analysis

Outside reviews, the Tools menu gathers utilities for any researcher: searching several databases, normalising bibliographies, verifying citations and working with a single article.

Verify citations before submitting

Citation verification is especially useful with bibliographies produced by AI tools: it detects references that do not exist, DOIs that point to another article and years or authors that do not match. Paste the bibliography, run it and review the red rows.

Article analysis and teams

Each analysed article can be shared with other users. The critical report exports to Word. Conversations are kept with the article.

Link to this chapter →

Article analysis

A tool unique to SysRevAI, outside the reviews: upload an article or your own manuscript and the AI writes a critical report, discusses it with quotes from the text and helps you improve the writing.

It lives under Tools → Articles. Each article is a separate project, with its document, the AI conversation, the critical report, related documents and its own team. AI must be enabled (Admin → Claude).

Creating a project

1. Tools → Articles → Upload an article.
2. Give it a title and upload the document as PDF or Word (.docx, .doc), up to 50 MB. The platform extracts its text.
3. Optionally add related articles or guidelines (for example the CONSORT statement or a reference study): the AI reads them alongside the main one.

Critical report

Press Critical report and choose the language (Catalan, Spanish, English or French). The AI reads the article like a demanding peer reviewer and returns:

• scores from 0 to 100 on five axes: methodology, clarity, novelty, evidence and handling of limitations;
• an executive summary;
• the devil’s advocate: the strongest objections a reviewer would raise;
• concrete recommendations for each section.

The report can be regenerated and exported to Word, on its own or with the conversation transcript.

Chat with citations

In the chat, ask anything about the article (“what is the sample size?”, “which limitations do the authors acknowledge?”). Answers quote the text of the article and of the related documents, so you can check them. The conversation is saved with the project.

Help writing your own manuscript

1. Open the Collaborative editor: the document opens in an editor with formatting, tables and images.
2. Select a passage (or nothing, to work on the whole document) and ask the Copilot for what you need: “rewrite this in a more formal tone”, “clarify this sentence”, “add a table with the results”.
3. The Copilot proposes the change and you accept or reject it.
4. Save named versions to go back to whenever you like, and download the document as Word.

Working as a team

From Team, invite co-authors by e-mail: if they already have an account they are added; otherwise they get an invitation link valid for 7 days. Everyone sees the document, the report and the conversation.

Link to this chapter →

AI: what it does, what it costs and how to control it

SysRevAI uses Claude language models (Anthropic) with your own key. AI assists; it does not decide. Here is what it does at each phase, how to enable it and how to keep the cost under control.

Enabling it

1. Create an API key in the Anthropic console.
2. Admin → Claude (AI): paste the key (stored encrypted), pick the default model and a monthly spending limit.
3. Switch each feature on or off: screening, extraction, summaries, chat, protocol, citations, Copilot.
4. Inside each review, the coordinator can switch AI off entirely.

What it does at each phase

Costs and limits

Every call is logged with the model, tokens and estimated cost in euros. In the review, the AI usage tab breaks it down by feature and person; under Admin → Claude (AI) you set the global monthly limit, and when it is reached the features stop without affecting the rest of the platform.

The AI-use statement

Each review automatically produces an AI-use statement: in which phases AI was used, with which models and how much. It exports to Word for the article’s methods section, as more and more journals require.

Link to this chapter →

Living reviews and the API

A living review stays current: saved searches re-run on their own and flag new references. The API lets other tools read the data.

Saved searches

1. Run an external search from the review (Search → Databases).
2. Press Save search: the query, filters and databases are stored.
3. Under Review → Saved searches, enable automatic re-runs (weekly or monthly).
4. The worker runs it again; new references come in marked "new since the last search" and you get a notification.

Read-only API

Under Profile → API you can create a personal token. With it, other programs can read your reviews, references and PRISMA as JSON.

curl -H "Authorization: Bearer YOUR_TOKEN" https://your-install/api/v1/reviews

Link to this chapter →

Updates, backups and maintenance

All maintenance happens in the panel: updating to the new version, making and downloading backups, putting the site in maintenance mode and watching the worker.

Updating

1. The platform checks sysrevai.com for a new version every 6 hours by itself. When there is one, administrators see a dark bar at the top of every page (“A new version of SysRevAI is available”) with an Update now button. The bar cannot be dismissed and goes away by itself once updated.
2. The button opens Admin → Maintenance with the check already done: the available version, what is new and the pre-flight checks. You can also go there yourself and press “Check for updates”.
3. If every check passes, press "Update now".
4. The platform switches maintenance mode on, backs the database up, downloads the package with your licence, verifies its integrity, swaps the code in (keeping .env, storage/ and public/uploads/), applies migrations and reopens the site.
5. If a migration fails, the previous code is restored and the backup is at hand. Each update is recorded with its log.

Backups

Admin → Maintenance → Full platform backups: every 2 hours the worker backs up the whole platform (database, PDFs, protocols, articles, credentials, avatars and .env) and keeps the 24 newest; older ones are deleted automatically. You can change the frequency and how many are kept, make one by hand and download them. To restore one, on the server run php bin/console.php backup:restore storage/backups/<file>.zip --yes (--with-env also restores .env). Download a copy off the server from time to time.

Maintenance mode

Shows a notice page to everyone except admins. Useful during migrations or server moves.

Worker and jobs

Admin → Maintenance → Scheduled tasks (cron) says whether the worker is running and when it last ran; if it is not, it gives the exact cron line for this server and the steps to set it up. Admin → Background jobs shows the task queue (bulk AI assessments, scheduled searches, builds).

Where everything lives

Exporting and importing data (portability)

Every user can download everything they created from Profile → My data: a ZIP file with their reviews (protocol, references, decisions, conflicts, extraction, risk of bias, GRADE, tags, comments and the AI usage log), the PDFs and the articles. The same file is imported into any other installation from the same screen.

• Imported reviews belong to whoever imports them; collaborators are matched by e-mail and, without an account, are created deactivated (they take no seat until an administrator switches them on).
• An import is all or nothing: if anything fails, nothing is saved.
• To import large files, raise upload_max_filesize and post_max_size in the server’s PHP settings.

Link to this chapter →

Support

Every trial account and every installation, Community and Institution, has a built-in support system: you ask the SysRevAI team for help with tickets, from inside the platform.

Who can use it

Other users (reviewers, invited members) see under Support who administers their platform, so they can contact that person directly.

Opening a ticket

1. Press the Support pill in the top bar, next to “User guide”.
2. Press New ticket. Write the subject, choose the type (technical problem, licence and installation, suggestion or other) and the priority, and describe what happens, what you expected and how to reproduce it.
3. If needed, attach files: up to 5 per message (screenshots, PDF, Office or LibreOffice documents, text, logs, CSV, RIS, BibTeX or ZIP), 5 MB each and 10 MB in total.
4. Send it. You get an e-mail when the SysRevAI team replies.

System information

Each ticket automatically carries the installation’s technical details: version and edition, PHP and database, missing extensions, server limits, number of users and reviews, cron job status and whether e-mail and AI are configured. That way the support team sees the environment without having to ask. It contains no passwords, keys or personal data. On the new-ticket page it is collapsed: open it with the arrow on the “System information” box if you want to check it.

Following the ticket

• SysRevAI’s replies appear in the ticket’s conversation, with any files attached to them.
• The Support pill shows in red how many tickets have a reply you have not opened yet; in the list they are marked “New reply”.
• You can reply, with files too, as often as needed. Replying to a resolved or closed ticket reopens it.
• When you no longer need it, press Close ticket.

Institution edition: installation on your servers

The Institution edition includes help installing SysRevAI on your organisation’s servers, coordinated with your IT department. To request it, write to us from Contact.

Link to this chapter →

The mobile app

SysRevAI has an app for iPhone, iPad and Android, on phones and tablets. Screen references, read the full text, resolve conflicts, comment with your team and use the tools, even with no connection.

One app serves every SysRevAI: the trial accounts on sysrevai.com and any institution’s own installation. You sign in with the same account as on the web, see the same reviews, and your decisions end up on the same server. What works better on a big screen (administering the platform) stays on the web, and the app takes you there when needed.

Installing it

1. On an iPhone or iPad, search for “SysRevAI” in the App Store.
2. On an Android phone or tablet, search for it in Google Play.

Connecting it to your server

What you can do

• Home: what is waiting for you in each review, and the “Continue screening” button.
• Screening by title and abstract and by full text, with the protocol’s terms highlighted, exclusion reasons, notes and the AI proposal when the server has AI.
• Article list: the button at the top of screening shows the articles still to screen and the ones you have screened. Tap one to screen it, or to change your decision.
• Full text: the PDF with the terms highlighted, search (with the magnifier, which tells you how many results there are) and zoom, and the protocol’s criteria beside it. When the AI has assessed the article, each criterion shows as met, not met or unclear.
• References: the review’s list with filters (status, text, tag, source) and actions on the ticked ones: screen them by title and abstract, tag them or delete them (the owner only). Also import a file from the phone (Files, iCloud, Drive, an attachment) or pasted text, with a preview; search for new references with the protocol’s filters and save the search; saved searches; and duplicates.
• Each reference’s page (from the list or from screening): the full text (open the PDF, upload one from the phone, replace it, delete it or search the open repositories now), the markers at each stage, and the AI summary, peer review and chat. In The review’s full texts, coverage and the background search queue. In conflicts, the AI’s proposed resolution.
• Managing the review: create a new one (the AI can fill in the form from the protocol document, PDF or Word), edit the protocol, the team (invite by e-mail, roles, blinding, who resolves conflicts), the settings (archive, the public read-only link, delete an archived review) and the coordination of each stage: start it, share the work out, and have the AI assess the queue and accept its most confident verdicts. Each person sees what their role allows, as on the web.
• Included studies, built for tablets first, with the study’s PDF or abstract beside the form: data extraction (the AI can fill your draft; the owner edits the template and approves extractions), risk of bias (RoB 2, ROBINS-I, Newcastle-Ottawa or JBI, with the traffic light, the summary plot and the AI’s proposal per domain) and GRADE with the summary of findings table. On a phone, a switch at the top moves between the form and the study.
• Results and reporting: the PRISMA diagram (and the editor to pin cells by hand), every file of the web’s Exports tab (downloaded to the phone, with the share sheet), the PROSPERO/OSF registration form with the AI’s draft, and the AI-use declaration. Comments can be deleted (your own, or any if you own the review).
• Conflicts, for whoever resolves them.
• Comments on each reference and on the review’s board. Typing @ suggests the team, and the person mentioned gets a notification.
• Alerts from the platform and push notifications.
• Tools: the same as on the web (below).
• Support, for administrators: questions to the SysRevAI team, with screenshots.

Working offline

• The app keeps a batch of references from your queue on the phone. Before a trip, tap “Download 200 references to work offline”.
• Each decision is saved at once and sent as soon as there is a connection; the top of the screen shows how many are waiting. Until it is sent, you can undo the last one.
• If another reviewer completed the reference while you were offline, the app tells you your decision did not count: the server always has the last word.
• At the full-text stage, the batch’s PDFs (the first 20) download with it and read offline too. Those already on the phone carry the “Saved for offline reading” mark.
• Comments, the AI proposal and the tools need a connection. Profile → Offline data shows how much space it all takes and lets you clear it.

On a tablet

• On an iPad and on Android tablets, navigation moves to a sidebar.
• Screening shows in two columns: the batch on one side and the reference on the other.
• At the full-text stage, the PDF sits on one side and the criteria, AI proposal, notes and decision on the other.
• With a keyboard attached: I includes, M marks as maybe and E excludes (as on the web) (then 1-9 picks the reason and 0 none).
• The tools show the list and the detail together. With the app in half the screen (Split View or Android windows) it goes back to the phone layout without losing what you were doing.

Notifications

The app tells you when you are invited to a review, references are assigned to you, there is a conflict, someone mentions you or work is returned to you. On the web, the “Mobile app” column of Profile → Notifications decides which ones you get. A notification only carries a generic sentence (“You have a new conflict to resolve”); when you tap it, the app asks your server for the details. Self-hosted installations send notifications through sysrevai.com with their licence, but without titles, names or decisions.

Tools

• Verify citations in CrossRef, OpenAlex and Europe PMC (up to 15 at a time): pasted, from a review, or by scanning a printed bibliography with the camera. The text is recognised on the phone itself, and you check it before verifying.
• Normalise citations to any of the 8 styles, to copy, share or add them to a review.
• Your latest verifications and normalisations stay under Recent, to open them again.
• Search references in the external databases, in your own references or with a clinical question, and add the ones you pick to a review; duplicates are caught.
• Article analysis: upload a PDF or Word file, generate the critical report, ask the AI about the article or improve a paragraph of your manuscript. Normalising, the report and the chat need the server to have AI set up (chapter 14).

Language, security and updates

• Language: the app follows the phone’s, and you change it in Profile → Language. It has every language of the platform, and a new one appears without updating the app.
• Devices: each phone has its own session. On the web, Profile → Mobile app lists your connected devices and lets you sign out one you have lost.
• Blind mode: it applies just as on the web. In a blinded review, the app does not receive the rest of the team’s decisions.
• Updates: the app updates itself in the background and uses the new version the next time it opens; the profile shows a notice to restart it straight away.
• Signing out deletes the saved batches and PDFs from the phone. Decisions not yet sent are lost, so do it with a connection.

Common questions

Link to this chapter →

Frequently asked questions

Short answers to the questions we get most often. If yours is missing, write to us.

Link to this chapter →

Good practices

Methodological and organisational recommendations we have seen make reviews faster and more defensible.

Before you start

• Close the protocol and register it (PROSPERO, OSF) before importing references. Changing criteria mid-screening forces you back.
• Run a pilot with 50 references and the whole team. Compare decisions, adjust criteria, then open the real phase.
• Define the exclusion reasons with the team: few, mutually exclusive and ordered from most to least restrictive.

During screening

• Short, regular sessions (30–45 minutes) keep consistency better than marathons.
• Review conflicts every week: settling them early recalibrates the team.
• Watch the kappa. Below 0.6, stop and clarify the criteria.
• Use "Maybe" deliberately: it is a reserve, not a way to postpone the decision.

With AI

• Start with prioritisation; automatic screening only for very low-scoring references and always with a verification sample.
• Always verify pre-extraction against the quoted passage.
• Include the AI-use statement in the article’s methods.

Security and data

• Two-step verification for everyone with an admin role.
• Check in Admin → Maintenance that the cron says “Running”: without it there are no automatic backups. Download a full backup off the server from time to time.
• Update when a new version is out: updates back up and roll back if needed.
• Do not upload PDFs you may not share with the team under your licence.

Link to this chapter →