Version: 1.0 | Author: Radosław Wojciechowski
Type: Standalone offline tool | Format: Single HTML file, no installation required
DOCTA is a standalone, offline tool for analysing software documentation and generating test artefacts using a local AI model (Ollama). No installation, no cloud connection, no subscription required — your documents never leave your computer.
- Accepts text, PDF (.pdf), Word (.docx), and image input (.jpg / .png / .gif / .webp)
- Generates:
- Functional requirements (R-01, R-02...)
- Test cases (TC-001 format) with categories: HAPPY PATH, NEGATIVE, BOUNDARY
- Ambiguities & gaps
- Questions for dev/analyst
- Coverage matrix (requirements → test cases)
- Exports to .txt, .md, .docx, .xlsx
- Batch mode — analyse multiple files, export all results to a single Excel file
- Analysis history stored in browser localStorage (up to 50 entries)
- Prompt templates for consistent output style across projects
- EN / PL interface and output language toggle
- 100% offline — no API keys, no accounts, no internet required after setup
Download and install from https://ollama.com/download.
Ollama starts automatically in the background after installation.
DOCTA runs in a browser and requires Ollama to allow cross-origin requests.
Windows (Command Prompt):
set OLLAMA_ORIGINS=* && ollama serveMac / Linux:
OLLAMA_ORIGINS="*" ollama serve
⚠️ Run this command every time you restart Ollama. Consider creating a startup script or shortcut.
Recommended for 8 GB RAM (text analysis):
ollama pull qwen2.5:7bFor image / screenshot analysis:
ollama pull llava:7bOpen docta.html in any modern browser (Chrome, Firefox, Edge). No web server needed.
The status indicator in the top bar should show a green dot and "Ollama running".
If it shows red, check that Ollama is running with CORS enabled (step 2).
DOCTA uses a two-panel layout: sidebar navigation on the left, main content area on the right.
| Section | Description |
|---|---|
| Analyser | Main analysis tool — single document |
| History | Past analyses stored in browser localStorage |
| Batch Analyser | Analyse multiple files in sequence |
| Settings | Ollama URL, default model, prompt templates |
| About | Version and technical information |
Three input modes are available:
- Paste text — paste plain text from any source (Confluence, email, Notepad, etc.)
- Upload file — drag and drop or browse; supports
.txt,.pdf,.docx - Image / screenshot — upload
.jpg,.png,.gif,.webp; requires a vision model (e.g.llava:7b)
Steps:
- Paste or upload your documentation
- Select output options (requirements, test cases, ambiguities, etc.)
- Select output language (EN / PL)
- Click Analyse
- Export results as
.txt,.md,.docx, or.xlsx
- Drag and drop multiple
.txt,.pdf, or.docxfiles - Select output options and language
- Click Analyse all
- Export all results to a single
.xlsxfile
- Every completed analysis is saved automatically
- Up to 50 entries stored; each shows date, model, token count, and document preview
- Click any entry to reload it into the Analyser
- Export full history to
.xlsx
Templates append extra instructions to every analysis prompt for consistent output style.
| Template | Description |
|---|---|
| API Testing | Focus on REST API; includes HTTP status codes, schema validation notes |
| Functional Spec | IEEE 830 style; hierarchical numbering; flags conflicting requirements |
| User Story | Given/When/Then format; focuses on acceptance criteria |
| GDPR / Security | Highlights personal data, GDPR risks, auth/authorisation test cases |
Custom templates can be created and saved in localStorage. Only one template is active at a time.
| Model | Size | RAM | Best for |
|---|---|---|---|
qwen2.5:7b (default) |
~5 GB | 8 GB | Text analysis — best quality/speed balance |
mistral:7b |
~4 GB | 8 GB | Fast, good English output |
llama3.1:8b |
~5 GB | 8 GB | Strong reasoning |
llama3.2:3b |
~2 GB | 4 GB | Fastest, lower quality |
phi3:mini |
~2 GB | 4 GB | Lightweight, quick |
llava:7b |
~5 GB | 8 GB | Image/screenshot analysis |
llava-phi3:3.8b |
~3 GB | 6 GB | Faster vision analysis |
moondream:1.8b |
~1 GB | 4 GB | Lightest vision model |
llava:13b |
~8 GB | 12 GB | Best vision quality |
Recommended for 8 GB RAM:
- Text:
qwen2.5:7b - Images:
llava:7borllava-phi3:3.8b
| Problem | Solution |
|---|---|
| Red dot — Ollama not reachable | Verify Ollama is running (ollama list); confirm CORS flag is set; check port 11434 is not blocked by firewall |
| Analysis stops or returns empty | Model may exceed available RAM — try llama3.2:3b or phi3:mini; split long documents into sections |
| Image analysis fails | Ensure a vision model is selected; pull it first with ollama pull llava:7b |
| PDF text not extracted | Scanned (image-only) PDFs cannot be read via Upload mode — use the Image tab instead |
| History not persisting | Do not use private/incognito mode; do not clear browser data between sessions |
All processing is performed locally via Ollama. No data is sent to any external server.
No API keys, no accounts, no internet connection required after initial Ollama setup.
DOCTA — Documentation Test Analyser | v1.0 | Author: Radosław Wojciechowski