Detect logical fallacies in articles and pasted text.
Self Hosted • Beautiful Interface • Multilingual
- What is Fallacy Finder?
- Why Fallacy Finder?
- Key Features
- Screenshots
- Quickstart with Docker
- Configuration
- How It Works
- Roadmap Ideas
- Contributing
- License
Fallacy Finder is a self-hosted web application for analyzing arguments and surfacing potential logical fallacies in a readable, visual way.
Paste in text or submit a URL, let the app extract and review the content, and then run it through a language model to identify possible fallacious reasoning. Results are displayed with highlighted passages, counters, expandable explanations, and a built-in library of fallacy definitions.
Fallacy Finder is designed for people who want something more useful than a generic chatbot answer. The project was made for students, educators and inquisitive minds.
Most commercial LLMs can identify a fallacy if you ask properly. Very few are built to make the process usable, reviewable, and repeatable.
Fallacy Finder focuses on:
- A better workflow
Inspect extracted text before analysis begins - A better reading experience
Highlighted passages, badges, and structured result views - A better reference experience
Built-in fallacy library with detailed explanations - Flexible deployment
Use local models via Ollama or cloud models via OpenAI - Accessibility for real users
Multilingual interface and polished UI
- Analyze web articles by URL
- Analyze custom pasted text
- Review extracted content before spending tokens or compute
- Highlight passages that contain likely fallacies
- Show fallacy counts, metadata, and result summaries
- Optionally include reasoning/explanations for each detected finding
- Ollama support for local/self-hosted models
- OpenAI support for hosted models
- Model-aware saved results and settings workflows
- Provider connection testing from the UI
- Built-in Fallacy Library
- Individual fallacy detail pages
- Related fallacies and classification metadata
- Useful for critical reading, teaching, and rhetorical analysis
- Dark, modern interface
- Multilingual UI
- Responsive layout
- Review step between extraction and analysis
- Clear alerts and settings flow
services:
fallacyfinder:
image: jlaroche/fallacyfinder:latest
container_name: fallacyfinder
ports:
- "8784:8080"
environment:
- SECRET_KEY=change-me-to-a-long-random-string
- OLLAMA_BASE_URL=http://localhost:11434
- OLLAMA_MODEL=phi4:14b
# Optional OpenAI support
- OPENAI_API_KEY=
volumes:
- ./instance:/app/instance
- ./logs:/app/logs
restart: unless-stoppeddocker compose up -dThen open:
http://YOUR-SERVER-IP:8784
Fallacy Finder is designed to work well in self-hosted environments. While we encourage experimentation with different models, after extensive testing we recommend using Microsoft's Phi-4 (14b) model with Fallacy Finder.
Common Configuration Options:
SECRET_KEY— Flask session secret (can be generated here)OPENAI_API_KEY— optional, only needed if using OpenAI
Extended Configuration Options
Options for HTTPS and Reverse Proxy support can be found in the .env.example file.
You can also tune additional settings from within Fallacy Finder's UI.
- Submit a URL or paste text
- Review extracted content before analysis
- Send the text to the configured LLM provider
- Parse structured fallacy results
- Display findings visually with highlights, metadata, and fallacy details
The goal is not to claim perfect philosophical certainty. The goal is to make argument analysis more transparent, inspectable, and practically useful.
Fallacy Finder is especially useful for:
- students learning argumentation and rhetoric
- teachers building classroom examples
- researchers working with discourse and persuasion
- journalists and readers evaluating public claims
- anyone who wants a practical tool for spotting questionable reasoning
- more base fallacies
- export/share analysis results
- compare outputs across multiple models
- increase multilingual support
- improved source extraction for difficult websites
- user accounts and saved analysis history improvements
Contributions are welcome!
Especially useful contributions include:
- UI and accessibility improvements
- translation improvements
- extraction improvements for tricky websites
- model prompt and parsing improvements
- documentation and test coverage
If you open an issue or PR, thank you — seriously.
Distributed under the GPL-3.0 license. See LICENSE for more information.
Built with 💞 by Jacques Laroche as a Project for the Center for Digital Curiosity - a multidisciplinary, student-focused labority built and operated in the heart Belgrade, Serbia.