AI Text Detector
Detect if text was written by ChatGPT, Claude, or a human. Our AI detector runs 100% locally in your browser using open-source neural networks. Private, secure, and free.
- 100% private & local
- No character limits
- Works offline
Ready to scan
Paste your text and click Detect. The model runs entirely in your browser.
- Runs in your browser
- No upload
- No sign-up
- Open-source models
- No character limits
Related tools
Free AI and developer tools that run in your browser. Nothing uploaded, ever.
How this compares to other AI detectors
Most AI detectors charge a subscription and upload your text to their servers. This one runs locally.
| Feature | TToolDeck | GPTZero | Originality.ai |
|---|---|---|---|
| Text never sent to a server | Yes | No | No |
| No account or email required | Yes | Partial | No |
| Runs via WebAssembly (ONNX) | Yes | No | No |
| Works offline after first load | Yes | No | No |
| No character limits | Yes | No | No |
| Cost | Free | Paid | Paid |
How it works
- 1
Paste your text
Paste an essay, article, or paragraph. For the highest accuracy, provide at least 100 words.
- 2
Download the model
The first time you use it, your browser downloads a 125MB open-source neural network. It is cached for future use.
- 3
Get the result
The model analyzes the text structure and outputs a probability score of whether it was written by AI or a human.
Everything you get for free
100% Private
Your text might contain confidential information. Because the model runs locally via WebAssembly, your data never leaves your device.
Open-Source Model
We use the RoBERTa-based OpenAI Detector model from Hugging Face, fine-tuned specifically to recognize LLM-generated patterns.
No Character Limits
Commercial detectors limit you to 1,000 characters on their free tiers. We process the entire text in your browser's memory.
Instant Analysis
Once the model is cached, analyzing a 5,000-word document takes less than a second.
Transparent Confidence
We show you the exact confidence percentage so you understand how strongly the neural network feels about its prediction.
Works Offline
After the first use, you can disconnect from the internet and the detector will continue to work perfectly.
What people use it for
Academic Integrity
Educators can screen student submissions to identify potential misuse of ChatGPT.
Content Verification
Journalists and editors can verify if a submitted article was written by a human or generated by AI.
SEO Auditing
Check if a guest post or backlink article was spun by an AI content farm before publishing it on your site.
Freelance Vetting
Ensure your writers are delivering original human-written content as promised, not LLM outputs.
Personal Curiosity
Test your own writing to see if your style closely mimics the structural predictability of AI models.
AI Model Testing
Prompt engineers can test how well their latest prompts bypass or trigger AI detector models.
Questions people ask
Are my prompts uploaded to a server?
No. ToolDeck uses the Transformers.js library and ONNX Runtime Web to download the model directly to your browser. Your text is processed locally and never leaves your device.
Why does it take so long the first time?
The first time you use the tool, your browser must download the 125MB neural network model from Hugging Face. After that, it is saved in your browser cache and loads instantly.
How accurate is this AI detector?
It uses the RoBERTa-base OpenAI Detector, which achieves roughly 95% accuracy on GPT-2 generated text. However, newer models like GPT-4 and Claude 3 can sometimes produce text that bypasses it. Treat the result as a probability, not absolute proof.
Does it detect ChatGPT-4 and Claude?
Yes, but with varying accuracy. Modern LLMs are trained to be more human-like and predictable. The detector looks for low perplexity (predictability) and low burstiness (uniform sentence length), which are common AI signatures.
Is there a character limit?
No. Since the model runs locally in your browser, the limit is only your device's memory. It can easily handle a 10,000-word document.
Can I use this for academic grading?
While it is a useful screening tool, AI detectors are known to produce false positives (flagging human text as AI). We highly recommend using this alongside other methods rather than as definitive proof of misconduct.
Why does it flag my human writing as AI?
If you write highly structured, grammatically perfect, and predictable text (like a legal document or academic paper), the model may flag it. AI models are trained on vast amounts of formal text, so they mimic this style closely.
Does this work offline?
Yes. Once the 125MB model is downloaded and cached in your browser, you can disconnect from the internet and the detector will continue to work perfectly.