Qualitative Data Analysis Software Compared

NVivo, ATLAS.ti, and MAXQDA are paid desktop programs for analyzing interviews, documents, and media. At list price, an NVivo researcher license costs $520 a year, an ATLAS.ti educational license costs $387 a year, and a student license for either costs $130 a year. Taguette and QualCoder are free, open-source alternatives for one coder or a small team. Potato is a free, self-hosted alternative to NVivo and ATLAS.ti for team coding, with a reliability sample assigned to several coders, Krippendorff's alpha per question, and REFI-QDA .qdpx exchange with the paid programs.

Qualitative coding assigns short labels, called codes, to passages of text such as interview transcripts or open-ended survey answers, and a codebook lists the codes and defines each code. An in-vivo code takes the name of the code from the participant's own words. Software built for qualitative coding is called CAQDAS, for computer-assisted qualitative data analysis software. Intercoder reliability measures how often two or more coders apply the same codes to the same material, usually as Cohen's kappa for two coders or Krippendorff's alpha for any number of coders. REFI-QDA is an open standard for moving a codebook (.qdc) or a whole project (.qdpx) between programs.

The tools below differ in price, in where your data is stored, in which reliability coefficient each tool reports, and in where AI-assisted coding sends your text. Qualitative research in Potato covers Potato's qualitative features in one place, and Survey Research Tools Compared covers survey software.

Which qualitative coding tools are worth comparing?

Prices come from each vendor's own store or pricing page in October 2026, in US dollars.

Dedoose charges individuals and small groups only for the months in which a user logs in. Usercall prices plans by a monthly credit balance shared between interviews and analysis.

Which tool for which task

The table pairs each kind of qualitative project with the tools built for that kind of project.

NVivo, ATLAS.ti, and MAXQDA appear in many rows because each program covers analysis by one researcher, reliability between coders, and project exchange in one product. Potato appears where a project assigns material to coders and reports agreement between the coders.

Free NVivo and ATLAS.ti alternatives

Three tools on this page cost nothing to license. Taguette is the simplest of the three, for highlighting and tagging documents on your own computer, on a server you run, or on app.taguette.org, and Taguette reports no reliability coefficient. QualCoder covers the most media types of the three, coding text, PDFs, images, audio, and video on one computer, with Cohen's kappa between coders and an optional AI agent that can use a local model. Potato is the free option built for a coding team, with a reliability sample assigned to coders automatically, Krippendorff's alpha with a bootstrap interval, and REFI-QDA exchange with NVivo, ATLAS.ti, and MAXQDA. None of the three matches every analysis feature of the paid programs, and what Potato does not do lists Potato's gaps.

Hosting and data location by tool

NVivo Desktop and MAXQDA store projects on your own computer. NVivo teams share projects through NVivo Collaboration Server or through NVivo Collaboration Cloud, an add-on at $99 a year. NVivo AI Cloud runs on Microsoft Azure, with data hosted in Europe, the United States, Canada, Singapore, or Australia. Lumivero plans real-time multi-user editing in AI Cloud for early 2027. MAXQDA TeamCloud gives each member a local copy of the project and exchanges the work through the cloud on a schedule, with one team lead and up to four members per license.

ATLAS.ti Desktop stores projects locally, and ATLAS.ti Web hosts projects on servers in Germany with real-time collaboration. Dedoose is a cloud application, and a project can be shared with any other Dedoose user at no additional cost. Usercall names Supabase, OpenAI, and Anthropic among Usercall's service providers and says that data may be processed across regions.

Taguette runs on your own computer, on a server you run, or on app.taguette.org, and several people can work on one project on a server. QualCoder opens a project on one computer at a time. A QualCoder team either passes the project file from one person to the next or codes copies separately and merges the copies into a master project. Potato runs on your own computer or your institution's server, and every coder works in a browser against the Potato server. Each coder's work is written to the study's output directory, and Potato sends text to an outside service only when the config names such a service, as with a hosted AI endpoint.

Intercoder reliability by tool

Each tool reports a different set of coefficients. A coefficient missing from a tool's set needs a separate calculation on the exported codes.

A dash means we found nothing for the row in the tool's own documentation, checked on 2026-10-28. ATLAS.ti's manual says that Cohen's kappa was left out on purpose, citing limitations in the literature such as kappa's restriction to two coders. If a cell is wrong, tell us.

Potato's reliability export writes one row per question, with the level of measurement set by the question type. Code choices from a list are nominal, and a multiselect codebook uses MASI distance. Two coders who chose overlapping sets of codes therefore get partial credit. The export does not cover highlighted passages. For passages, the admin dashboard at /admin/iaa reports token-level kappa, span F1, a character-level Krippendorff's alpha, and Mathet's gamma. Agreement for Spans explains how the span measures differ, and the agreement calculator computes alpha and kappa from a CSV of codes exported by any tool.

AI-assisted coding and where the text goes

NVivo, ATLAS.ti, MAXQDA, and Usercall send text to a hosted language model for the AI features of each product. MAXQDA, QualCoder, and Potato can use a model on your own computer or server instead. The destination matters when a consent form or an ethics approval limits where interview text may be sent.

• NVivo. AURA, the assistant in NVivo AI Cloud, summarizes sources, suggests initial topics, and answers questions about the project with citations to the sources. Lumivero states that uploaded data is not used to train AI models and is not retained after a task. NVivo Desktop has no AI features.
• ATLAS.ti. AI Coding, AI Summaries, and AI Code Suggestions run on OpenAI's systems, and ATLAS.ti Desktop users choose the United States or Europe as the processing region. ATLAS.ti states that OpenAI does not train on data sent through the OpenAI API.
• MAXQDA. Cloud AI Assist runs through MAXQDA's server in Frankfurt and an AI model server in the EU, and MAXQDA states that the data is never used for training. Since MAXQDA 27, AI Assist Premium can use a model on your own computer or server through an OpenAI-compatible connection such as Ollama, llama.cpp, or LM Studio.
• Dedoose. We found no AI-assisted coding in Dedoose's help center.
• Usercall. An AI moderator conducts voice interviews, and Usercall codes transcripts and imported open-ended answers line by line into themes that researchers review and refine. Usercall states that customer data is not used to train AI models.
• Taguette. Taguette has no AI features.
• QualCoder. QualCoder 4.0 has an AI agent that can read the project, add codes in a sandboxed mode, or edit codes and codings after you confirm each destructive change. The agent uses OpenAI, Google Gemini, OpenRouter, Microsoft Azure, Blablador, or a local model through Ollama. A local model indexes the documents, and only selected passages of about 500 characters go to a cloud provider.
• Potato. With ai_support, a language model offers hints, highlights keywords, and suggests codes on each item, and the coder decides. Setting endpoint_type to ollama or vllm keeps the model on your own server, and openai, anthropic, and gemini send each item to the named provider. AI Support covers the configuration.

Moving projects between tools with REFI-QDA

A REFI-QDA project file moves sources, codes, and codings between programs, and each program documents which of the program's own features stay behind. NVivo imports and exports .qdpx with sources, codes, cases, memos, and coding, and leaves surveys, queries, and matrices behind. ATLAS.ti and MAXQDA import and export .qdpx, and ATLAS.ti notes that a QDPX bundle contains only the items every participating program shares. Dedoose exports a project as .qdpx. QualCoder imports and exports .qdpx, and QualCoder's author does not yet guarantee full compliance with the standard. Taguette exports a project's codebook as a REFI-QDA .qdc file and a whole project as a Taguette database.

Potato imports a .qdpx file as a new project and exports a coded project back to .qdpx:

potato import -i interviews.qdpx -o interviews-study
python -m potato.export -c interviews-study/config.yaml -f qdpx -o ./export/

The codebook keeps the hierarchy of codes, and each coding keeps the coder who made the coding. ATLAS.ti allows only one level of subcode, so for a deeper codebook add --option flatten_subcodes=true to the export. The option moves every code to the top level and renames each code Parent > Child. Importing a Project covers the offsets and the warnings the importer writes.

Team coding with a reliability sample in Potato

The config below sets up a team to code interview excerpts about access to a health clinic. Every excerpt is coded once. For reliability, a second coder codes a 20% sample of the excerpts drawn within each site. The codebook is fixed for this round, so coders apply the agreed codes and cannot add new ones. Memos are shared with the team, and cases group the excerpts by participant, with the site as a case attribute.

annotation_task_name: Clinic access interviews
task_dir: .
output_annotation_dir: annotation_output/
data_files:
  - data/excerpts.json
item_properties:
  id_key: id
  text_key: text
user_config:
  allow_all_users: true
  users: []
 
# One coder per excerpt, and two coders on a 20% sample drawn within each site
num_annotators_per_item:
  default: 1
  overlap_sample:
    fraction: 0.2
    count: 2
    stratify_by: site
    seed: 7
 
# Coders apply the agreed codes and cannot add new ones during this round
codebook_mode: fixed
 
annotation_ui:
  memos: true
  visibility: shared
 
cases:
  enabled: true
  key: participant_id
  attributes: [site]
 
annotation_schemes:
  - annotation_type: span
    name: quotes
    description: Highlight each passage that shows one of the codes.
    codebook: true
    labels: [access barriers, cost concerns, provider trust, administrative burden]
  - annotation_type: multiselect
    name: themes
    description: Which codes apply to this excerpt as a whole?
    codebook: true
    labels: [access barriers, cost concerns, provider trust, administrative burden]

Each excerpt in data/excerpts.json has the fields the config names:

{"id": "p01-03", "text": "I couldn't get an appointment for almost three weeks, and the clinic was nowhere near a bus line.", "participant_id": "P01", "site": "rural"}

The config leaves QDA Mode off. QDA Mode assumes one analyst over the whole corpus and opens the codebook to edits, and the study above needs a fixed codebook and two coders on the sample. Both schemes draw on one shared codebook. The quotes scheme records the passages for a quotation report. The themes scheme records the codes for each excerpt as a whole, and the reliability export scores the themes answers with Krippendorff's alpha. Run the three exports once coding is done:

python -m potato.export -c config.yaml -f reliability -o ./export/
python -m potato.export -c config.yaml -f quotation_report --option include_memos=true -o ./export/
python -m potato.export -c config.yaml -f qdpx -o ./export/

The first command writes reliability.csv and reliability.md with alpha, a 95% bootstrap interval, and percent agreement per question. The second writes one row per highlighted passage, with the quote, the quote's offsets, the excerpt, and the coder, followed by the memos. The third writes a .qdpx file for colleagues who continue in NVivo, ATLAS.ti, or MAXQDA. QDA Mode covers the single-analyst setup with in-vivo coding, Content Analysis covers codebooks loaded from a file, and Adjudication covers settling the disagreements.

What Potato does not do for qualitative research

• Codes text only. The codebook, in-vivo coding, and REFI-QDA exchange work on text. Recordings become text first with potato transcripts --transcribe --diarize, a command that runs Whisper on your own computer. NVivo, ATLAS.ti, Dedoose, and QualCoder code audio and video directly.
• Imports text sources only from REFI-QDA. Pictures, PDFs, audio, and video in a .qdpx file are not imported, and a project with no text sources arrives empty, with a warning.
• Leaves memos and cases out of the .qdpx export. The export contains the sources, codes, codings, and coders, and the memos go out in the quotation report.
• Two analysis tables. Potato's admin dashboard has a code co-occurrence matrix and a table of codes by case attribute. NVivo has queries and coding matrices, and further analysis of Potato's coding runs on the exports in R, Python, or another QDA program.
• Scores reliability on whole items in the export. The reliability export covers questions with choices, scales, and numbers, and agreement on highlighted passages is on the admin dashboard only.
• Runs no interviews. Usercall conducts AI-moderated interviews, and Potato starts from transcripts or text you already have.
• Needs a server. Someone installs Potato with pip and runs the server, on a laptop for one coder or on a shared machine for a team. A Hugging Face Space is a hosted option, and Taguette's app.taguette.org needs no installation.

Checked against each project's own documentation and pricing pages in October 2026.

Frequently asked questions

Which qualitative coding software calculates Krippendorff's alpha?

Of the tools on this page, ATLAS.ti and Potato calculate Krippendorff's alpha. ATLAS.ti Desktop reports the Krippendorff alpha family alongside percent agreement and the Holsti index, after the coders' projects are merged. Potato's reliability export reports alpha per question with a 95% bootstrap interval. NVivo, Dedoose, and QualCoder report Cohen's kappa, and MAXQDA reports Brennan and Prediger's kappa. For codes exported from any tool on this page, the agreement calculator computes alpha from a CSV.

What is the best free qualitative data analysis software?

Taguette, QualCoder, and Potato are the free, open-source tools on this page, and the best of the three depends on the project. Taguette suits highlighting and tagging documents, on your own computer or on a shared server. QualCoder suits one researcher coding text, PDFs, images, audio, and video, with AI assistance from a cloud provider or a local model. Potato suits a team that splits the material among coders, needs Krippendorff's alpha on a reliability sample, and exchanges projects with NVivo or ATLAS.ti users as .qdpx files. All three tools cost nothing to use.

Further reading

• Qualitative research in Potato, for the qualitative features in one place
• QDA Mode
• Content Analysis
• Code Interview Transcripts With Two Coders and Report Alpha, for a team coding project from the codebook to alpha, adjudication, and a QDPX file
• Importing a Project
• Inter-Annotator Agreement, for choosing and reading a coefficient
• Agreement for Spans
• Adjudication and Disagreement
• Survey Research Tools Compared
• Annotation Tools Compared

References

1. Jurgens, D., Chen, M., & Iyer, L. (2026). Potato 2.0: A Comprehensive Annotation Platform with AI-in-the-Loop Support. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations), 374-386. https://aclanthology.org/2026.acl-demo.37/