What SciSpace is and the problem it addresses
SciSpace is a research support platform operated by PubGenius Inc., previously known as Typeset. It brings together publication discovery, assisted reading, study comparison and preparation of academic materials. It is best understood as a scientific workspace rather than a single database or an independent AI model.
A user can start with a question, bibliographic references or documents. The output may be a paper list, an explanation linked to a passage, an extraction matrix or a draft. These are products at different stages of development: a retrieved paper still needs reading; generated prose still needs checking.
Reference documentation: About SciSpace.
Origins and development from Typeset
The official history traces the origins to 2015, with Sai and Shanu Kumar, focused on publication formatting. The SciSpace name was announced on 11 May 2022, retaining the team and extending the scope beyond writing and layout.
The Product Hunt launch record documents AI Academic Writer in September 2024, Deep Review in February 2025, SciSpace Agent in August 2025 and BioMed Agent in December 2025. These are presentation dates on that platform, not necessarily the first availability of each component. The sequence illustrates the transition from assisting an editorial task to linking search, analysis and material production.
Reference documentation: About SciSpace · Introducing SciSpace — Revolutionizing research workflows end-to-end · SciSpace launches on Product Hunt.
Search and the provenance of literature
Literature Review supports paper discovery and comparison through configurable columns. The April 2024 data source statement names Semantic Scholar, OpenAlex, Google Scholar and other repositories. The presence of a bibliographic record does not mean SciSpace can access the full text.
Precise work requires separating three sets: retrieved publications, documents actually read and studies ultimately included. A broad occupational noise search might return interventions, exposure studies and papers on effects; combining them in one conclusion would obscure the question each design addresses. The table should show that classification before synthesising findings. A paper missing from an answer does not establish that no relevant literature exists.
Reference documentation: SciSpace Literature Review · SciSpace Data Sources.
PDF reading and contextual explanations
Chat with PDF supports questions about documents, summaries and explanations of passages, tables or formulas. Passage references and notes help readers return to the original. The practical value is making a difficulty locatable, rather than replacing all reading with a conversation.
A useful request separates description from appraisal: first ask which procedure the authors used; then locate the limitations they acknowledge; finally develop your own assessment. If an answer explains a statistical term, check that it applies the explanation to the analysis actually used. Keep the document name and version with the note, especially when both a preprint and a final article exist.
Reference documentation: SciSpace Chat with PDF.
Comparison matrices and data extraction
Extract Data offers custom columns, section references and exports such as CSV, Excel and bibliographic formats. This turns reading into a reusable structure. Each row should represent a clear unit, such as a study; each column should represent a variable defined before extraction.
Sabentis proposes separating value, unit, period and evidence location. “Exposure” is not a precise enough column if some publications describe intensity and others duration. Empty cells should distinguish unreported information, an inaccessible document and failed extraction. When two articles use the same sample, recording that relationship prevents counting them as independent observations. An exported sheet allows those decisions to be reviewed outside the conversation.
Reference documentation: SciSpace Extract Data.
Writing, references and editorial integrity
AI Writer provides suggestions, completion and citation support. The bibliographic generator can prepare in-text citations and full references from source details. SciSpace also offers an AI text detector with sentence analysis and scores.
These functions address different problems. A correctly formatted reference may point to a paper that does not support the sentence; fluent prose may change the authors’ level of certainty. Sabentis proposes checking each substantive claim against its source and documenting human and tool contributions. A detector score should not be used alone as proof of authorship or misconduct. Academic quality depends on provenance, reasoning and method as well as writing.
Reference documentation: SciSpace AI Writer · SciSpace APA Citation Generator · SciSpace AI Detector.
Agents, biomedical research and deliverables
The enterprise offering describes workflows combining extraction, synthesis, reports and presentations. BioMed Agent adds a focus on biomedical research, including genomics, pharmacology and multi-omics analysis. This specialisation describes the advertised product; it does not demonstrate suitability for deciding on an individual person or occupational exposure.
Before delegating a sequence, define its checkpoints: confirm the question before searching, accept the study set before synthesising and review findings before generating slides. Otherwise, an early mistake can reappear in a very convincing presentation. For OSH, a bounded reading and comparison workflow is generally more useful than an open request to “solve” a prevention problem.
Reference documentation: SciSpace Enterprise · SciSpace BioMed Agent.
Access, plans and resource consumption
The March 2026 guide distinguishes Basic, Premium, Advanced and Max. Credits apply to agent tasks; standalone tools are treated separately. The guide covers monthly credit expiry, concurrency limits and fair use for tiers advertised as unlimited.
Evaluating a subscription therefore requires estimating complete tasks, not just submitted questions. A useful pilot records documents processed, usable output and the review required. It should also test what happens if the balance is exhausted partway through a task. Current charges, included features and team conditions must be checked when purchasing: an older Writer page may describe a different combination of services.
Reference documentation: SciSpace Agent Credit Pricing and Usage Guide.
Browser, mobile and API access
The Chrome extension brings explanations of text, mathematics and tables into browsing. The mobile page links to an App Store application. Check each client separately: a feature visible on the web does not imply identical mobile availability.
The 4 November 2026 announcement presents the API as early access with a waiting list. It distinguishes Search API for reference retrieval from Task API for research workflows. This profile does not present it as generally available. For a future integration, a team should retain the query, date, article identifiers and output version; automation that saves only the final prose loses the traceability needed to update it.
Reference documentation: SciSpace Chrome Extension · SciSpace Mobile App · SciSpace API: Automate Research Paper Search & Literature Reviews.
Data, confidentiality and enterprise controls
The linked policy, effective from April 2024, covers data categories, providers and other third parties, advertising purposes and retention criteria. Enterprise advertises SSO, role-based controls and SOC 2 Type 2 certification. These are provider statements; their implementation and contractual scope have not been audited here.
For organisational use, separating public literature from internal material avoids treating both as equivalent. Before uploading restricted documents, clarify permissions, subprocessors, deletion and the terms applicable to the plan. An occupational health record does not become suitable for uploading simply by removing a name: other details may identify the person. The pilot described in this profile uses only a public article.
Reference documentation: SciSpace Privacy Policy · SciSpace Enterprise.
Academic reception and evaluation criteria
The Melbourne Institute of Technology library includes SciSpace in its literature review tools guide, updated in January 2026. It describes multilingual search, matrices and PDF reading. Inclusion documents academic interest and introductory guidance; it is not an independent accuracy trial.
To evaluate organisational usefulness, Sabentis proposes a test set of familiar papers: one with complex tables, another with missing information and another with qualified conclusions. Compare data retrieval, fidelity of explanations and total review time. The meaningful saving comes after errors have been corrected, not when an answer first appears. This profile does not report an evaluation conducted by Sabentis.
Reference documentation: Typeset/SciSpace — Melbourne Institute of Technology Library.
Applications in occupational safety and health
In OSH, SciSpace can support critical reading of evidence. The following uses are Sabentis proposals, not certified prevention functions or implementation results.
- Industrial hygiene: locate how exposure is defined, which instrument is used and what period it covers. The aim is to understand study conditions before comparing them with those at a workplace.
- Ergonomics: separate laboratory trials, observations in actual workplaces and intervention evaluations. Improvement in a controlled test does not automatically describe adoption throughout a working day.
- Technical training: prepare explanations of difficult concepts with the original passage available to the instructor. Material should retain conditions and exceptions that alter its meaning.
A useful reading record distinguishes observed results, author interpretation and the team’s judgement. That separation allows a decision to be revisited when new evidence appears without reconstructing the entire conversation. The output supplies documentation for prevention work; it is not in itself a workplace assessment.
Hypothetical case: reading an occupational noise study
A practitioner needs to understand a public noise paper before including it in a professional development session. The objective is not to decide whether a site meets a limit, but to explain accurately what the study measured and how far its conclusions extend.
- Prepare the document: save the reference, version and access date. Check that methods, tables and notes are readable. Record the questions the reading must answer in advance.
- Locate the method: request the measured quantity, instrument, duration, population and circumstances, with supporting passages. Keep retrieved information separate from details that are not reported.
- Check units: inspect headings and footnotes directly. Do not treat an average, a maximum and a personal measurement as interchangeable.
- Cross-check the synthesis: select one table and verify its values manually. Ask about a variable the paper does not report to check whether that absence is acknowledged.
- Write the teaching record: separate design, findings, limitations and open questions. Every conclusion intended for training should have a location in the paper.
If a chart and the text seem to disagree, retain both and resolve the difference before completing the record. Acceptance means another person can reconstruct the explanation from the document. No workplace measurements are invented and findings are not extrapolated to individual workers. This is an illustrative case, not an executed product validation.
Interpretation limits and reproducibility
An explanation may be clear while losing a decisive condition. Tables with multiple headers, formulas and scanned documents need particular checking. When an answer does not reveal its documentary basis, it should remain a reading hypothesis rather than an extracted fact.
A systematic review requires a defined question, a reproducible strategy, selection criteria and study appraisal. Chaining searches and summaries does not by itself establish that method. Distinguishing a primary article, a review and a subsequent correction also matters: adding their mentions together is not the same as accumulating independent evidence.
For future updates, retain the query, date, filters, included list and reasons for exclusion. Human review should focus on points that change the decision, rather than being confined to polishing the final report.
When to choose it and complementary tools
SciSpace fits situations where understanding documents and moving from reading to comparison is the bottleneck. Elicit can complement structured extraction; Scite enables examination of subsequent citation context. These address different questions rather than establishing a universal ranking.
The decision should consider corpus type, export requirements, data controls and checking effort. If the task is simply to archive validated references, adding generation may complicate a straightforward process. If it requires explaining dense methods to an interdisciplinary team, contextual reading may offer a specific benefit worth testing.
Frequently asked questions
Are SciSpace and Typeset unrelated products?
No. Typeset is the previous name in the product’s development. Date older instructions, because editing features and the current agent do not necessarily share the same plans.
Is a cited answer sufficient for citing a scientific result?
Open the article and check that the passage supports the claim. The citation helps with that check but does not replace it.
Can a generated matrix be considered complete?
Only after checking variables, units, missing information and the provenance of each relevant value. A tidy table may conceal duplicate studies or documents that could not be read in full.
