scienceOS
scienceOS is a German-built AI research agent that answers scientific questions with citations drawn from an index of 230 million papers, chats with up to 4,000 uploaded PDFs, and organises literature inside shared research projects.
What is scienceOS?
scienceOS is an AI research agent for scientists, published by scienceOS UG (haftungsbeschraenkt), a small company based in Dresden, Germany. It is delivered as three connected surfaces: an AI Science Chat, an AI PDF Chat and an AI Project Manager, all reached through a single web application.
What sets it apart is the corpus it answers from. The internal scienceOS paper index is built on more than 230 million research papers taken from the Semantic Scholar index, together with 800 million full-text snippets and 2.4 billion reference pairs, and it is refreshed weekly. Answers are grounded in the full text of open access papers and in the abstracts of closed access ones, and every answer cites its sources.
The terms of use describe four ways to query it. Science Chat searches the internal index. File Chat searches the documents you have uploaded into a conversation. Library Chat searches the personal library you have built. A deep research mode combines all three for a broader pass. Alongside the chat sit three distinct tools: a Library holding up to 4,000 sources, a Citation network that draws the links between papers, and a Consensus analysis that examines what citing papers actually say about the works they cite.
Document handling is the second pillar. You can upload up to 4,000 PDFs, capped at 128 pages on the Free and Angel plans, 1,024 pages on Team and Institution, and 80 MB per file. A PDF+ extraction available on paid plans reads scanned documents as well as figures and tables. Sources come in from PDFs, local folders, Zotero on desktop and web, or DOIs, and can be exported again.
Projects turn all of this into shared work: you add sources, invite colleagues, annotate findings, and decide whether attached PDFs may be copied by the people you invite. A free model is available to everyone, while a better model, deep skills and bulk import belong to the paid tiers. Everything runs in the browser, on servers the publisher places inside the European Union, with a business offer adding separated EU instances and optional self-hosting in a customer's own Azure environment.
What it does
- Ask a scientific question and receive an answer that cites the papers it is grounded in
- Chat with up to 4,000 of your own PDFs at once and extract data across them
- Run a deep research pass across the paper index, your chat files and your library at the same time
- Import references from Zotero, local folders, DOIs or raw PDFs, and export them again
- Draw citation networks to explore how papers reference one another
- Generate comparison tables and diagrams from the literature
- Create shared project workspaces and invite colleagues to contribute sources and findings
When to use scienceOS / When not to
A quick filter to help you decide if scienceOS is the right fit.
When to use scienceOS
- Academic researchers and postdocs who run literature reviews and need every claim tied back to a citable source
- PhD candidates and advanced students mapping an unfamiliar field before writing a proposal or a paper
- Laboratory and research groups that want a shared workspace for sources, chats and annotated findings
- Universities and research institutes that need institutional login, usage-based pricing and separated EU server instances
- GDPR-sensitive organisations that require EU data residency, a Data Processing Agreement on request and a contractual no-training guarantee
When not to use scienceOS
- Anyone under 18: the terms of use forbid access to the application below that age
- Companies expecting the advertised prices to apply, since those rates cover academic, non-profit and personal use only and industry pricing is quoted by email
- Developers who want to call the tool from their own code: there is no API and the terms explicitly prohibit automated or programmatic access
- Teams that need to process special categories of personal data, which the terms of use prohibit users from entering
- Users who need a mobile or offline companion, as scienceOS runs in the browser only with no iOS, Android or desktop application
How to use scienceOS
A typical end-to-end flow, from setup to results.
- Open app.scienceOS.ai and create an account with an email address and password, or through a single sign-on provider
- Start on the free plan, which needs no payment details, and ask a first question in the chat bar
- Pick the model you want for the task and attach AI Actions or specific sources to frame the answer
- Read the cited sources behind the answer and click a citation to open the citation network around it
- Upload PDFs, or import them from a local folder, from Zotero on desktop or web, or by DOI, to build your library
- Switch on the deep research mode when a question needs the index, your chat files and your library at once
- Create a project, add the sources that matter, annotate the findings and invite colleagues to contribute
- Export your sources whenever you want to continue the work in another reference tool
- Upgrade through the Angel programme page if you need the best model, figure extraction and bulk import
- Delete the account from the settings section and confirm with the verification code if you want to stop
Pros & Cons
Pros
- Every answer cites its sources, and the corpus behind them is named openly: the Semantic Scholar index, refreshed weekly
- The no-training promise is contractual, not just marketing: the terms warrant that neither Input nor Output feeds large language model training
- The permanent free plan is unusually generous, covering the science chat, text PDF chat and citation networks
- Data residency is announced across the European Union, the publisher is EU-based, and a DPA is offered on request
- Every subprocessor is named individually in the privacy policy, with its address and legal basis
- The academic entry price is low at 7.25 euros per month, and non-profit partnerships add member discounts
- Development is visibly active and documented, with public release notes and reference import and export for interoperability
Cons
- There is no API, and the terms of use explicitly forbid accessing the application automatically or programmatically
- No mobile or desktop application exists, so the tool is browser-only with no offline access
- The published prices only cover academic, non-profit and personal use; anything commercial requires a quote
- The 6.00 euro Team rate is per user and requires at least six Angel licences, so the real entry cost is six times that figure
- The Institution plan has no public price at all: a one-off setup fee and computation credits are negotiated by email
- No retention period is published for chats, uploaded files or libraries; only the 30-day server log window is quantified
- No Data Protection Officer is named, no interface language is declared, and both published email addresses are obfuscated on the page
Pricing & Plans
A permanent free plan is available at 0.00 EUR and gives access to the fast model, text-only PDF chat and citation networks. The lowest paid entry point for an individual subscriber is the Angel plan at 7.25 EUR per month, excluding taxes. The Team plan is quoted at 6.00 EUR per user per month excluding taxes, but requires a minimum of six Angel licences, so its real starting cost is higher. The Institution plan is priced on request and combines usage-based pricing with a one-off setup fee. All advertised prices apply exclusively to academic, non-profit and personal use; industry pricing must be requested by email.
- 0.00 EUR - fast model with a usage limit
- text-only PDF chat
- citation networks
- 7.25 EUR per month plus taxes - everything in Free plus unlimited fast model
- best model with a usage limit
- PDF chat covering text and figures
- deep skills
- bulk import of 100 files
- 6.00 EUR per user per month plus taxes
- from six Angel licences - everything in Angel plus multi-tiered discounts
- unlimited best model
- extra-long PDFs
- bulk import of 4
- 000 files
- annual invoicing
- enhanced data security
- custom price - everything in Angel plus usage-based pricing
- unlimited best model
- extra-long PDFs
- bulk import of 4
- 000 files
- annual invoicing
- enhanced data security
- institutional login
Data, GDPR & hosting
A consolidated view of how scienceOS handles your data.
GDPR overview
GDPR implementation is concrete and above average. The publisher is established inside the European Union, so no Article 27 representative is required and none is named. The site claims a fully GDPR-compliant research agent, states that models, chats, uploaded files and user data sit on servers within the European Union, and offers a Data Processing Agreement on request to Team and Institution subscribers, together with support for EU AI Act compliance requirements. The privacy policy, effective 21 November 2025, names a legal basis for every processing operation, lists each subprocessor with its address, documents TLS/SSL encryption and privacy by design, and enumerates the Articles 15 to 21 rights plus the parallel Swiss data protection rights. Two gaps: no Data Protection Officer is named, and the only contact address is the general one.
Who owns the data?
The controller is scienceOS UG (haftungsbeschraenkt), based in Dresden, Germany. The terms grant you a non-exclusive, non-transferable and non-sublicensable right to use the application; in return you grant the vendor the right to use your Input for the sole purpose of producing Output, even where that Input is protected by copyright. You warrant that you hold every right and permission over what you upload, and you stay responsible for how the Output is used. Uploaded PDFs are stored encrypted in your account and reachable only by you unless you add them to a shared project. Nothing in the terms assigns ownership of the Output either way.
Reuse rights
Nothing in the terms of use restricts what you do with the answers the tool produces, and no clause claims ownership over them, so you may reuse your Output without asking permission. Three limits still apply. You may not rely on the Output as a sole source of truth: the vendor states it can be incomplete, inaccurate or offensive and requires human review before you use or share it. You may not use Output relating to an individual for any purpose carrying a legal or material effect on that person, such as credit, education, employment, housing, insurance, legal or medical decisions. And the application itself may not be resold, transferred, rented, modified or accessed programmatically. On the vendor's side, the terms warrant that neither your Input nor your Output nor any other user information is used by scienceOS or its subcontractors to train large language models, and the site offers a template for disclosing AI use in a manuscript.
Data retention & training
Hosting summary
The publisher states that its AI and machine learning models, user chats, uploaded files and user data are all located on servers within the European Union, and describes strict EU data residency and processing controls. No individual hosting country is named, so the European Union is the finest granularity actually published. Business customers are offered separated EU server instances for their organisational data, and an opt-in to self-host scienceOS inside their own Azure Cloud environment. The privacy policy names ALL-INKL as the IT infrastructure provider, with a Data Processing Agreement supplied by that provider, and confirms TLS/SSL encryption in transit; uploaded PDFs are stored encrypted and reachable only by their owner unless shared in a project. Some processing still leaves the European Union: the payment providers Stripe and Paddle are United States companies, and Google services are used, with transfers covered by the EU-US Data Privacy Framework, adequacy decisions or standard contractual clauses. The website itself is served through a content delivery network, which is routing rather than storage.
Things to keep in mind
Risks and trade-offs to weigh before adopting scienceOS.
- Cited answers feel authoritative, but the publisher warns that Output can be incomplete, inaccurate or offensive and requires human review before use or sharing
- Leaning on an AI agent for literature discovery can quietly erode the habit of reading widely and judging sources yourself, which is the core skill the tool is meant to serve
- Academic integrity is at stake: many journals require AI use to be disclosed, and the publisher supplies a disclosure template for exactly that reason
- You grant the publisher the right to use your Input to produce Output, including copyright-protected material, so check what you may lawfully upload before you upload it
- Entering special categories of personal data under Article 9 GDPR is prohibited by the terms, and Output about an individual must not drive legally or materially significant decisions
- No retention period is published for chats, files or libraries; on termination the publisher may irretrievably delete everything and leaves the backup to you
- Commitments are longer than the monthly headline suggests: three months minimum on Team, one year on Institution, invoiced in advance and payable within seven days
Setup & Integrations
Technical difficulty
Very easy for an individual. There is nothing to install: scienceOS runs in the browser, you create an account with an email address and a password or through a single sign-on provider, and the free plan opens immediately with no payment details required. Importing a library from Zotero, a local folder or a list of DOIs takes a few clicks. The effort only rises at the institutional end, where the Institution plan involves institutional login, a one-off setup fee and computation credits agreed by email, and where self-hosting the platform in a customer's own Azure Cloud environment is the single genuinely technical option on offer.
Deployment
Integrations
Behind scienceOS
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Which papers can scienceOS actually search?
Is there a genuinely free version?
How much does the paid version cost?
Does scienceOS train its models on my data?
Where is my data stored?
Is there an API I can call from my own code?
What are the limits on PDF uploads?
Is there a mobile app, and is there a minimum age?
Can I get a Data Processing Agreement?
How do I close my account and what happens to my data?
Should you pick scienceOS?
scienceOS is a deliberately narrow tool, and that is its strength. It does not browse the open web; it answers from a named corpus of scientific publications and shows you the papers behind every claim. For a researcher who spends their days moving between a literature index, a pile of PDFs and a reference manager, the combination of grounded chat, a 4,000-document library and shared project spaces covers most of the daily loop in one place.
The privacy posture is more solid than the marketing language alone would suggest. The no-training promise is written into the terms of use as a warranty rather than left in a FAQ, EU data residency is stated plainly, subprocessors are listed one by one with addresses, and a Data Processing Agreement is available to team and institutional subscribers. For a European lab weighing an AI tool against its own compliance rules, that is a short and checkable list.
The gaps are equally clear. There is no API, and programmatic access is contractually forbidden, so this will not slot into an automated pipeline. There is no mobile application. No retention period is published for the content you upload, only for server logs. And the advertised prices are conditional: they hold for academic, non-profit and personal use, while anything commercial goes through a quote, and the headline team rate needs six licences before it applies.
The publisher is a young, small German company that says openly that an early fundraising attempt did not work out and that it turned to a paid plan instead. The release notes show steady, documented work regardless. For academic literature work under European data rules, scienceOS is a serious candidate; for anyone needing automation, mobility or firm retention commitments, it is not yet.
- Choosing a selection results in a full page refresh.
- Opens in a new window.