The answer in two sentences. Stay with ATLAS.ti if you need Windows, a web version, shared projects or a project exchange with colleagues on other platforms. Consider Liminon if you work alone on a Mac and your ethics protocol says recordings and transcripts must not be sent to an AI provider. Facts below were read on 2026-10-16.
Side by side
| Liminon | ATLAS.ti | |
|---|---|---|
| License model | One-time: $129 early-bird, $179 standard, $79 student; free during the beta (pricing) | Flat-fee licences per the home page; multi-user, course and non-profit options. Amounts not verified |
| Where the AI runs | On your Mac, through an optional local model in Ollama; suggestions stay pending until you accept them | OpenAI GPT models via API; you choose US or Europe processing; opt-in, extra fees may apply |
| Platforms | macOS 14+, Apple Silicon only | Windows, Mac and Web with any license; projects move between them |
| Transcription | On-device speaker-separated transcription; the first run downloads the models once | AI Auto Transcription, 30+ languages, automatic speaker detection |
| Team work | One person per project folder (.qproj); share by handing over the folder or exports | Asynchronous shared coding on desktop; cloud storage and sharing |
| Importing another tool's project | Not yet (REFI-QDA on the roadmap) | Not verified on the pages we read |
| Data retention statement | Local files; no cloud copy | Encrypted in transit and at rest (AES-256); retention length not stated |
| Free trial | Free during the beta, no account | Offered; length not stated on the page |
What ATLAS.ti does well
- Three platforms, one project. Its pages say you can use Windows, Mac and Web with any license and exchange projects between them. Its desktop page describes native interfaces on each system.
- Built-in transcription. AI Auto Transcription is listed with 30+ languages and automatic speaker detection; the page claims interviews are processed 20x faster than manual methods, a vendor figure we have not tested.
- Analysis tools. A network editor, word frequency tables and code co-occurrence analysis are listed on the desktop page.
- An open door for other AI tools. A new ATLAS.ti MCP Server lets a compatible AI tool work with your Desktop projects, which suits people who already run their own assistants.
- A clear AI data statement. The security page names OpenAI as the provider, offers US or Europe processing, says submitted data is not used to train models, and describes a data processing agreement with standard contractual clauses.
The AI question, read closely
ATLAS.ti's AI Coding, AI Summaries and code suggestions use OpenAI's GPT models through an API. The same page says processing happens on OpenAI systems and trusted service providers' systems worldwide, that limited authorized personnel and contractors may view content for abuse or security investigations, account support, legal compliance or fine-tuning (unless you have opted out), and that AI features are opt-in with possible additional fees. None of that is unusual for a cloud AI service, and it is stated openly. It does mean a protocol that bars sending transcripts to a third party cannot use these features as they stand. Retention length is not stated on the page we read.
When to choose ATLAS.ti
- You need to work on Windows, in a browser, or share projects across all three.
- Your institution holds a campus or multi-user license; the site lists course concurrent licenses and non-profit discounts.
- You want AI coding and your protocol allows a vetted cloud provider under a data processing agreement.
When Liminon is the better fit
- Your recordings are covered by an NDA or consent form that rules out sending them to an AI provider. Liminon's transcription, speaker separation and optional AI (a small local model through Ollama) run on your Mac; the only network use is the one-time model download and the update check you start.
- You want to pay once. Liminon is $179 ($129 early-bird, $79 student) with updates through v1.x. ATLAS.ti's home page speaks of one flat fee but we could not read an amount, so we do not compare figures.
- You want every AI step on record. Suggestions remain pending until you accept them, and the methods report lists what you accepted and rejected.
What you give up by switching
Windows and web access, shared projects, and the network editor. Liminon is Apple Silicon only (macOS 14+), has no live collaboration, and cannot import an ATLAS.ti project yet; REFI-QDA import is on the roadmap. English interviews work best.
Moving over
Export transcripts from ATLAS.ti and bring the original audio or video into a new Liminon project. Because projects do not transfer, the cleaner switch is at the start of a study, not the middle. Our comparison of twelve tools lists the alternatives if neither fits.
Sources
- AI data security and privacy, ATLAS.ti Desktop, students and education (all read 2026-10-16).
FAQ
Does ATLAS.ti use my data to train AI models?
Its AI security page says none of your data is used to train or improve AI models, and that OpenAI does not use API data to train its models. It also says content may be viewed for fine-tuning unless you have opted out, so read the page against your protocol.
Can ATLAS.ti AI run offline?
Not as described on its security page, which says AI requests go to OpenAI. Liminon's optional AI runs on a local model on your Mac.
Is Liminon an ATLAS.ti alternative for a dissertation?
If you work alone on an Apple Silicon Mac with mostly English interviews, yes, with the limits on this page. See also our dissertation page.
Can I import an ATLAS.ti project into Liminon?
Not yet. REFI-QDA import is on the roadmap and not in this build.
Try Liminon free during the beta
Transcribe, code, and analyze interviews on your Mac.
macOS 14+ · Apple Silicon (M1 or later) · Free during beta