Open Source Local-First Free

Qualitative AI Labeling Support

AI-Assisted Qualitative Analysis.
Done Right.

QUAILS brings systematic, unit-by-unit text analysis to qualitative researchers — with confidence scores, full prompt control, and a rigorous human-in-the-loop calibration workflow. Not a chatbot. A research instrument.

Download Free Read the Docs
localhost:5173 — QUAILS Analysis Builder

Screenshot: QUAILS visual flow canvas — showing an analysis pipeline with Open-Coding, Theme Identification, and Category Builder blocks connected on the drag-and-drop editor

5
Analysis Block Types
8+
File Formats
5
LLM Providers
100%
Local Option
MIT
Open Source License
The Problem

Chatbots Weren't Built for Qualitative Research

When you ask a chatbot to code your data, it chats — it doesn't analyze. There's no guarantee every unit gets reviewed. No confidence score per decision. No audit trail. No way to calibrate it against your own judgement. Results are inconsistent across sessions and impossible to replicate.

  • No systematic coverage — some units get skipped or lumped together
  • No confidence measurement — you can't tell when the AI was uncertain
  • No calibration — there's no way to align the AI with your coding schema
  • No replicability — the same prompt returns different results each time
  • No data custody — your transcripts go to someone else's servers
QUAILS is Different
  • Every unit is processed in order — guaranteed exhaustive coverage
  • Logprob-based confidence score on every single label decision
  • Per-block calibration with Cohen's Kappa before you run full analysis
  • Deterministic settings (temperature, seed) for reproducible results
  • Ollama option runs entirely on your machine — zero data leaves your desk
  • Full prompt customization and preview — you control exactly what the AI sees
  • Structured reasoning follow-up on every decision — auditable output
Key Features

Everything a Qual Researcher Actually Needs

Built around how rigorous qualitative coding actually works — not how a chatbot handles a question.

Systematic & Exhaustive
Every text unit is analyzed in sequence, with no skipping or batching common with AI tools. You know with certainty your whole corpus is reviewed. Each decision carries a logprob-based confidence score (0–100%) so you always know how certain the AI was.
Visual Pipeline Builder
Design multi-round analysis workflows on a drag-and-drop canvas. Connect blocks in any order — open-coding → theme identification → category building — with human review checkpoints built in between.
Three Analysis Modes
Open-Coding Analysis assigns free-text labels. Rubric-Based Scoring rates every unit against your criteria. Theme Identification runs constant comparison across coded units. Use them individually or chain them together.
Human-LLM Alignment
Before running full analysis, label a random sample yourself and compare against the AI with Cohen's Kappa. Iterate your prompts, re-run, and track improvement across saved test history — all within the app.
Full Prompt Control
Edit every system prompt, unit introduction, context framing template, and reasoning follow-up from the Default Prompt Editor. Preview the exact message structure the AI will receive before you run a single unit.
Complete Data Privacy
With Ollama, your data never leaves your machine. No internet required for analysis. No data is ever sent to the QUAILS developer. Cloud providers are optional and you are in full control of what you share.
Workflow

From Corpus to Codes in Five Steps

QUAILS structures the full qualitative analysis process — from document upload through calibration, analysis, and reporting.

1
Upload Your Corpus

PDF, DOCX, CSV, TXT, HTML, Excel — individual files or entire folder collections

2
Design Your Pipeline

Connect analysis blocks on the visual canvas, set unit type and prompts

3
Calibrate First

Human-label a random sample, compare with the AI, refine prompts until Kappa is acceptable

4
Run Full Analysis

Every unit processed systematically with confidence scores and reasoning logged

5
Review & Export

Browse results in the Data Browser, inspect in the Document Viewer, generate your report

localhost:5173 — Alignment & Calibration

Screenshot: Alignment panel — showing a sample of human-labeled units alongside LLM labels, with Cohen's Kappa score and per-label comparison table

Analysis Modes

Flexibility for Every Qual Approach

Whether you work inductively, deductively, or both — QUAILS has a block type for it. Chain them together for multi-round grounded theory workflows.

🏷️

Open-Coding Analysis

Two-step LLM process per unit: first a TRUE/FALSE relevance check, then a free-text label assignment. Both steps carry confidence scores from logprob extraction. A reasoning follow-up records why the AI labeled each unit.

Perfect for inductive coding where you want the AI to surface themes from the data rather than apply a pre-defined schema.

Inductive Free-text labels Confidence threshold Reasoning log
📊

Rubric-Based Scoring

Score every text unit against multiple independent rubric items in a single pass. Define the title, description, and scoring criteria for each rubric item — the AI scores, reports confidence, and justifies every score.

Ideal for deductive analysis, theory-driven coding, or structured content analysis where you already know what dimensions to measure.

Deductive Numeric scores Multi-rubric Justification
🔗

Theme Identification

Uses constant comparison — each labeled unit is compared against a growing theme list. If it matches an existing theme, it's assigned. If not, the AI generates a new theme title. Accepts input from one or more upstream blocks.

Built for grounded theory workflows where themes emerge iteratively from open codes generated in prior analysis rounds.

Grounded theory Constant comparison Multi-input Emergent themes
💡
Chain them together. Run Open-Coding on your corpus → feed labels into Theme Identification → feed themes into Category Building — all on one canvas, all resumable.
⚙️
Custom Prompt block. Need something none of the above covers? The Custom Prompt block lets you write any LLM task and direct it at any upstream output.
Rigorous by Design

Built for Research, Not Conversations

Every design decision in QUAILS prioritizes the things that matter in research: replicability, procedural consistency, and a defensible audit trail.

  • Deterministic by default — temperature=0, fixed seed, so the same prompt returns the same label on the same unit every time
  • Thinking mode disabled — reasoning tokens are suppressed so logprob confidence scores reflect only the answer, not the AI's internal deliberation
  • Per-unit isolation — no conversation history bleeds between units; each analysis call is fully independent
  • Configurable thresholds — set the minimum confidence required to accept a label; uncertain decisions are flagged separately
  • Version-controlled projects — every project has its own git history so you can trace every change to your analysis configuration
  • Checkpoint & resume — analysis can be stopped and resumed at any point; no work is lost
localhost:5173 — Activity Log

Screenshot: Live activity log showing unit-by-unit analysis output with labels, confidence scores, and reasoning for each processed text unit

Data Privacy

Your Data Stays With You

When you run QUAILS with Ollama, your documents never leave your machine. No internet connection is required for analysis. Full data custody. No data is ever sent to the QUAILS developer under any circumstances.

Feature Ollama (Local) Cloud Providers
Your documents leave your machine Never Yes — sent to provider
Internet required for analysis No Yes
Cost per analysis run Free (hardware only) API usage charges apply
Model selection Any Ollama-compatible model Latest frontier models
IRB / HIPAA / FERPA compatible Yes — fully local Depends — check provider policy
Data sent to QUAILS developer Never Never
Research compliance note: Before using cloud providers with research data, consult your IRB requirements and applicable regulations (HIPAA, FERPA, GDPR). When in doubt, use Ollama for fully local, offline analysis. Anonymous, opt-in crash reports are the only data that ever reaches the developer — and only with your explicit permission each time.
File Formats

Works With Your Documents

Upload individual files or entire folder collections. QUAILS parses every format into a four-level hierarchy (Document → Section → Paragraph → Sentence) and lets you choose the unit of analysis per block.

PDF DOCX TXT Markdown HTML CSV XLSX XLS
📋
Spreadsheet files (CSV/Excel) are analyzed row-by-row, with column selection — ideal for interview transcripts or survey data stored in tabular form.
AI Providers

Your Choice of Model

Start locally with Ollama — no API key needed. Switch to a cloud provider when you need a larger model. Change providers between projects without restarting the app.

Local Ollama Any Ollama model · hardware-aware recommendations
Cloud OpenAI GPT-4o and all GPT models
Cloud Anthropic Claude family
Cloud Google Gemini · Azure OpenAI
🏛️
Funding Acknowledgment
QUAILS was initially supported by the National Artificial Intelligence Research Resource (NAIRR) Pilot, grant #NAIRR240415, through the National Science Foundation (NSF) and OpenAI.

Ready to Start Coding Systematically?

QUAILS is free, open-source, and runs on your own computer. Download it, connect Ollama, and run your first analysis in under 10 minutes.

Download QUAILS Free Read the Documentation