Gatsbi
Gatsbi is an AI research assistant that generates original hypotheses, drafts complete academic manuscripts with verifiable citations, runs systematic reviews and meta-analyses, and writes patent disclosures. Built for researchers, engineers and R&D teams, on desktop and web.
What is Gatsbi?
Gatsbi is a research automation platform published by Gatsbi AI Limited, a Hong Kong company. It presents itself as a next-generation research assistant rather than a writing helper, and it is organised around three agents plus a patent mode.
Gatsbi Innovator handles ideation. It runs a systemic analysis of the problem, covering system and supersystem, components, substance-field relations, engineering parameters and contradictions, then returns 10 to 20 inventive hypotheses, each carrying a novelty score, a feasibility assessment and real references.
Gatsbi Writer handles drafting. It first classifies the type of research, whether methodological, experimental, clinical, case study or mixed methods, then produces a complete manuscript: abstract, introduction, literature review, methodology, results, discussion and conclusion, with in-text citations and a bibliography in APA, IEEE, Harvard, Chicago (NB) or AMA style. Since version 3.0, a Deep Research Agent plans the tasks, searches the literature, gathers and filters evidence from the web and synthesises it.
Gatsbi Reviewer handles evidence synthesis: searching and screening studies, extracting effect sizes, confidence intervals and sample sizes, running fixed-effect or random-effects meta-analyses with forest plots, funnel plots, I2 and Q statistics, and producing risk-of-bias or GRADE-style assessments along PRISMA lines. A separate Patent mode turns an idea or earlier material into a structured invention disclosure in any of 11 languages, while a paid Humanizer plugin rewrites academic prose paragraph by paragraph using Plugin Credits.
Under the surface, Gatsbi is an agentic workflow orchestrating several models, described as versions of ChatGPT and a Qwen/DeepSeek hybrid, whose detailed implementation the publisher declares confidential intellectual property. On the desktop app you can plug in your own OpenAI, Anthropic or Google API key, while the Hybrid option comes with the subscription and its tokens are covered. Distribution is a desktop application for Windows 10/11 and macOS on Apple Silicon, capped at three device activations, plus a web version reserved for Pro subscribers. The interface exists in 11 languages: the underlying reasoning stays in English and outputs are returned in the language of your input. NLTK and Pandoc are declared as open-source components. The publisher claims more than 100,000 researchers across 150 countries and 300 disciplines, and states plainly that Gatsbi does not submit academic work on behalf of users and does not guarantee publication outcomes.
What it does
- Generate 10 to 20 research hypotheses from a single topic, with novelty scores, feasibility assessments and real references
- Draft a complete academic manuscript, from abstract and introduction through methodology, results, discussion and conclusion
- Screen and synthesise literature for a systematic review, then run fixed-effect or random-effects meta-analyses
- Produce forest plots, funnel plots, I2 and Q statistics and publication bias checks, with risk-of-bias or GRADE-style appraisal
- Create figures, system diagrams, data tables, charts and LaTeX equations inside the manuscript
- Add in-text citations and a bibliography in APA, IEEE, Harvard, Chicago (NB) or AMA style, then export to Word, LaTeX or Markdown
- Write a structured patent disclosure in any of 11 languages, from an Innovator idea or from your own earlier material
When to use Gatsbi / When not to
A quick filter to help you decide if Gatsbi is the right fit.
When to use Gatsbi
- Academic researchers and PhD candidates who need to travel the whole distance from a research question to a submitted manuscript
- Postdoctoral researchers and research fellows running systematic reviews or meta-analyses along PRISMA and Cochrane lines
- R&D engineers looking for structured ideation and novelty-scored hypotheses on a concrete technical problem
- Corporate innovation and IP teams preparing invention disclosure drafts before handing them to a patent attorney
- Research groups and enterprises documenting innovation for grant applications, policy briefs or R&D tax credit files
When not to use Gatsbi
- Students hoping to hand in generated text as their own work: the terms forbid plagiarism, fabrication and falsification, and Gatsbi never submits anything on your behalf
- Anyone under 18, who is explicitly excluded by the privacy policy
- Developers who need a public API or programmatic access, since no API and no developer documentation is published
- Mobile-first users: there is no iOS or Android app, and the publisher warns that mobile browsers suspend long-running jobs
- Anyone looking for a permanently free tool or a long evaluation window, as the trial lasts one day and blocks document export, meta-analysis and plugins
How to use Gatsbi
A typical end-to-end flow, from setup to results.
- Create an account or sign in to your User Space on gatsbi.com
- Open the web version, or download the desktop app for Windows 10/11 or macOS on Apple Silicon and sign in there
- Activate the one-day free trial from the screen offered; no payment method is required and nothing is charged when it expires
- On desktop, choose your engine: Hybrid is included in the subscription, while the OpenAI, Anthropic and Google options need your own API key
- In Gatsbi Innovator, enter a topic or research problem such as battery life extension, then use Expand, Explain or Regenerate on the 10 to 20 scored ideas returned
- In Gatsbi Writer, paste a research note, describe a method or drop in your results; Gatsbi classifies the research type, then drafts the manuscript
- In Gatsbi Reviewer, enter a topic, choose a systematic review or a meta-analysis and apply an optional year filter
- Select or deselect the studies, upload your own PDFs or import references from Zotero, then check and edit the extracted data before generating
- Switch to Patent mode to turn an Innovator idea or earlier material into a disclosure, choosing the output language
- Export to Word (.docx), LaTeX or Markdown, with citations in APA, IEEE, Harvard, Chicago (NB) or AMA style
Pros & Cons
Pros
- Covers the whole research chain in one product: ideation, literature search, drafting, systematic review, meta-analysis and patent disclosure
- Generates technical material that the competitors named on its own comparison table do not produce: figures, system diagrams, data tables, charts and LaTeX equations
- References are presented as real and verifiable, drawn from academic databases rather than invented
- Explicit commitment not to train, fine-tune or improve any AI model on customer content, and desktop history that stays on your device outside the trial
- Choice of AI engine, including your own OpenAI, Anthropic or Google keys, with an 11-language interface and outputs returned in the input language
- Unlimited monthly usage advertised, with no hidden fees and cancellation at any time
- Entry price of USD 19.99 per month, or USD 159.99 per year, sits at the low end of the tools Gatsbi compares itself with, and the one-day trial asks for no payment method
Cons
- The free trial lasts a single day and blocks document export, meta-analysis and plugins
- No permanent free plan: the homepage shows a Free Plan card at USD 0 per day, but the pricing page and clause 6.1 of the terms list only the one-day trial and the paid plans
- No public API and no developer documentation, so nothing can be automated from outside the app
- No mobile application, and the publisher itself warns that mobile browsers suspend long-running jobs
- The OpenAI, Anthropic and Google engines on desktop require your own API key, a Tier 2 key or above being recommended, with token costs on top of the subscription
- GDPR is never mentioned in any document, no data protection officer or EU representative is appointed, no data processing agreement is published, and the governing law is that of Hong Kong
- Purchases are final and non-refundable in principle, with a 14-day request window and no guarantee of acceptance; the three-device limit is enforced through de-identified device identifiers, and monthly plugin credits do not roll over
Pricing & Plans
Gatsbi does not document a permanent free plan. Access is sold by subscription, and the lowest paid entry point is USD 19.99 per month with the Monthly Pro plan; Yearly Pro costs USD 159.99 per year, advertised as a 33% saving. A one-day free trial is available beforehand, with no payment method required and no charge on expiry, but it excludes document export, meta-analysis and plugins. Payment is taken by recurring card through Stripe or another processor offered at checkout, or by WeChat Pay as a non-recurring payment in RMB only, with 13% VAT added. Plugin Credits form a separate currency: 2,000 are granted each month with a subscription and do not carry over. Under clause 7.2 of the terms, purchases are final and non-refundable, although a refund request may be submitted within 14 days without any guarantee that it will be accepted.
- shown on the homepage only
- and understood to designate the one-day trial
- since neither the pricing page nor the terms describe a permanent free tier
- Monthly Pro
- USD 19.99 per month
- full features with a monthly subscription
- Yearly Pro
- USD 159.99 per year
- full features with a yearly subscription
- advertised as a 33% saving
- Both Pro plans include unlimited generation of hypotheses
- manuscripts
- patent disclosures
- systematic reviews and meta-analyses
- the OpenAI
- Anthropic and Google options with your own API key
- the Hybrid option with tokens covered
- the desktop app with three device activations
Data, GDPR & hosting
A consolidated view of how Gatsbi handles your data.
GDPR overview
The word GDPR appears nowhere on the site or in the legal documents, and Gatsbi never claims compliance: the privacy policy speaks only of applicable data protection law. Concrete elements are nevertheless present. Section 15 lists legal bases: contractual necessity, legal obligations, legitimate interests and consent. Section 7.3 offers restriction, objection, portability, withdrawal of consent and complaint to a supervisory authority, but only where applicable law provides these rights. Section 8 says international transfers rely on lawful mechanisms such as Standard Contractual Clauses, and section 22 promises breach notification within statutory deadlines. Section 18.2 confirms that no data protection officer and no EU representative have been appointed, only that contact details will be published if the law requires it. No data processing agreement is published. Requests go to helpdesk@gatsbi.com; the controller is Gatsbi AI Limited, Hong Kong.
Who owns the data?
Clause 8.2 of the terms is explicit: you retain full ownership of all content you generate using the software, meaning text, data, files and every output. Gatsbi AI Limited owns the software itself, its AI and front-end functionality, excluding user content, third-party content and open-source components such as NLTK and Pandoc. Where that content sits depends on the edition: on the paid desktop app your history stays locally on your device and never reaches the publisher's servers, while on the paid web version it is stored on Gatsbi's servers so you can open, edit and export it. API keys you supply are stored locally by default and are not collected.
Reuse rights
You may reuse everything Gatsbi produces without asking permission: ownership stays with you, so drafts, figures, tables and reviews can be edited, published or filed as you see fit, subject to your own academic and professional obligations. The publisher states that it does not use user queries, AI-generated outputs, research interests descriptions or critical exception logs to train, fine-tune or improve any AI model. Queries are nonetheless forwarded to the third-party AI providers you select or enable, among them OpenAI, Anthropic, Google, xAI, Alibaba Cloud and Transparent AI services, and your research interests description is sent to Transparent AI services to build recommendations. During the one-day trial, queries and responses are retained and analysed in aggregated or de-identified form for reliability, quality, abuse detection and debugging. Outside the trial, local desktop history is subject to no statistical analysis and no manual review, and web history is not used for product analysis unless you explicitly authorise it in a support ticket. De-identified device identifiers are collected to enforce the three-device limit.
Data retention & training
Hosting summary
Data may be stored on servers provided by Amazon Web Services in London (United Kingdom) and by Alibaba Cloud in Singapore and the United States (Silicon Valley), the exact location varying with configuration, user location, availability and operational needs. Requests handled through Alibaba Cloud large-model APIs may be processed in Singapore, the United States (Virginia), China (Beijing, North China 2) or Germany (Frankfurt), depending on endpoint, model and routing. Outside the trial, the desktop application keeps working history locally on the device rather than on the publisher's servers. Encryption in transit and at rest, access controls and monitoring are announced, while the policy states plainly that no absolute security can be guaranteed. The certifications quoted, ISO 27001 and SOC 2 for AWS and ISO 27001, SOC 1/2 and CSA STAR for Alibaba Cloud, belong to those hosting providers: Gatsbi AI Limited claims none of its own. Payment details are collected directly by the payment processor and never stored by the publisher. The governing jurisdiction is Hong Kong.
Things to keep in mind
Risks and trade-offs to weigh before adopting Gatsbi.
- The publisher disclaims any warranty of accuracy or reliability for generated content, so every claim, figure and reference has to be checked by a human before it is used or submitted
- Leaning on the tool to produce work you could not defend erodes the very skills a thesis or a review is meant to build, and the terms forbid plagiarism, fabrication and falsification outright
- Patent disclosures are drafts intended for review by a qualified patent professional: they are neither a filing nor legal advice, and the terms exclude any professional advice
- Your prompts leave the machine even in private desktop use, since they are forwarded to third-party AI providers that apply their own policies
- The three-device limit is enforced through de-identified device identifiers with no opt-out, and granted plugin credits expire at the end of each cycle
- Money is hard to get back: purchases are final in principle, the refund window is 14 days with no guarantee, and WeChat Pay is non-recurring, in RMB only, with 13% VAT added
- Personal data is deleted or irreversibly anonymised after 12 consecutive months without a login once access has expired, so an unattended account can quietly lose its archive; no data protection officer or EU representative is appointed to turn to
Setup & Integrations
Technical difficulty
Low. Onboarding takes three steps: create an account or sign in, launch Gatsbi, activate the one-day trial. The web version needs no installation and is reserved for Pro subscribers; the desktop app is a standard download for Windows 10/11 or macOS on Apple Silicon, with three device activations per account. No development skill and no API integration are involved. The only real friction sits on desktop, where the OpenAI, Anthropic and Google engines need your own API key, Tier 2 or above being recommended; the Hybrid option needs none. Zotero import is optional, and mobile browsers are discouraged.
Deployment
Integrations
Supported languages
Behind Gatsbi
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Gatsbi.
Frequently asked questions
Who publishes Gatsbi?
How much does Gatsbi cost, and is there a free version?
What is limited during the free trial?
Which AI model does Gatsbi run on?
Are the references real, or are they hallucinated?
Is my content used to train AI models?
Is my activity logged?
Where is my data hosted?
Is there a mobile app or a public API?
Can I submit the generated text as it is?
Should you pick Gatsbi?
Gatsbi aims higher than the writing assistants it is usually compared with: it wants to be an AI co-scientist covering the full cycle, from the first hypothesis to a systematic review, a finished manuscript or a patent disclosure. That end-to-end coverage is its real distinction, and it is backed by outputs most rivals do not generate at all, namely figures, system diagrams, data tables, charts and LaTeX equations. The publisher insists that references come from academic databases and are verifiable, commits explicitly to never training any model on customer content, and keeps desktop history on your own machine outside the trial. At USD 19.99 per month, the entry price is modest for the scope.
The reservations are just as concrete. The trial lasts a single day and withholds exactly the features a buyer would want to test: document export, meta-analysis and plugins. The Free Plan card on the homepage has no counterpart in the pricing page or the terms, so budget on a paid subscription. There is no public API, no mobile app, and the desktop OpenAI, Anthropic and Google engines expect your own API key and its token bill. On the legal side, GDPR is never named in any document, no data protection officer or EU representative has been appointed, no data processing agreement is offered, and disputes fall under Hong Kong law with purchases final by default.
For a researcher, a doctoral candidate or an R&D team that already knows how to judge a draft, Gatsbi is a serious accelerator. It is not a shortcut: the publisher itself requires human verification of everything it produces, and the responsibility for what you sign remains entirely yours.
- Choosing a selection results in a full page refresh.
- Opens in a new window.