Frequently Asked Questions

AI Troll is a mobile app that lets you run AI language models directly on your phone or tablet. All processing happens locally on your device—no internet connection required after downloading a model.

Yes, AI Troll is free to download and use. The AI models available through HuggingFace may have their own licenses, which are free for personal use in most cases.

Only for downloading AI models. Once a model is downloaded, you can use AI Troll completely offline.

Yes! AI Troll is designed with privacy as the top priority. All conversations happen locally on your device. No data is collected, no analytics, no tracking. Your conversations never leave your phone.

No. All AI processing happens 100% on your device. Your conversations are stored only in local app storage and are never transmitted to any server.

No. AI Troll does not require any account, registration, or sign-in.

None. AI Troll does not collect any personal data, usage analytics, or any other information.

AI Troll downloads models from HuggingFace, a popular repository of open-source AI models. You can also import your own GGUF models from your device storage.

AI Troll supports GGUF (GPT-Generated Unified Format) models. This format is optimized for efficient inference and includes everything needed to run the model.

Model sizes vary from several hundreds of MB to 10GB or more, depending on the model architecture and quantization level. Smaller quantized versions (Q4, Q5) are recommended for mobile devices.

For most phones, we recommend starting with smaller quantized models (Q4_K_M or Q5_K_M) in the 1-3GB range. These offer a good balance between quality and performance.

Yes. Go to Model Management and tap the delete option for any model you want to remove. This will free up storage space on your device.

Yes! You can import any GGUF model file from your device storage using the "Import Local Model" feature.

AI processing is computationally intensive. Larger models require more processing time. Try using a smaller/more quantized model, closing background apps to free up RAM, and ensuring your device isn't overheating.

This depends on the models you download. A typical setup with 2-3 models might require 3-10GB of storage.

AI processing is CPU-intensive, which does use more battery than typical apps. For extended sessions, we recommend keeping your device plugged in.

AI responses should be treated as suggestions, not facts. AI models can generate inaccurate, incomplete, or misleading information. Always verify important information from reliable sources.

No. You should NOT rely on AI responses for medical, legal, financial, or other professional advice. Always consult qualified professionals for such matters.

AI language models work by predicting the most likely next words based on their training. They don't actually "know" facts—they generate plausible-sounding text. This means they can confidently state incorrect information.

Yes! In Settings, you can adjust:

• Context Limit: How much conversation history the AI can remember (based on your device's RAM)

• Response Length: Maximum tokens per AI response

• Temperature: How creative/random vs. focused the responses are

• Top-K/Top-P: How many token choices the AI considers

• Repetition Penalty: Reduces repetitive responses

If the AI produces content you find inappropriate or offensive, you can flag and hide it directly in the chat. Long-press on any AI message to access the flag option. Flagged content will be hidden from view.

The model may be too large for your device's available memory. Try:

1. Close other apps to free up RAM

2. Download a smaller or more quantized version of the model

3. Restart your device and try again

Try these steps:

1. Check your internet connection

2. Ensure you have enough free storage space

3. Check if "WiFi Only Downloads" is enabled in Settings (downloads won't start on mobile data)

4. Try a smaller model first to test if downloads work

5. Note: Downloads continue in the background even if you close the app—you'll get a notification when complete

You can always delete chats in the list of chats, or clear the app's data from Android Settings.

RAM requirements depend on the model file size and quantization type:

By Model File Size:

- 1-2GB models: 4GB RAM minimum, 6GB+ recommended

- 2-4GB models: 6GB RAM minimum, 8GB+ recommended

- 4-6GB models: 8GB RAM minimum, 10GB+ recommended

- 6GB+ models: 12GB+ RAM (high-end devices only)

By Quantization Type: 

Quantization determines how the model is compressed: 

- Q4_K_M (highest compression): ~1-3GB files → 4-6GB RAM needed

- Q5_K_M (balanced): ~2-4GB files → 6-8GB RAM needed

- Q6_K (less compression): ~3-5GB files → 8-10GB RAM needed

- Q8_0 (minimal compression): ~4-8GB files → 8GB+ RAM needed

Quick Guide:

- 4-6GB RAM phone: Start with Q4_K_M models under 2GB

- 6-8GB RAM phone: Q4_K_M or Q5_K_M models (2-4GB)

- 8GB+ RAM phone: Q5_K_M or Q6_K models (up to 6GB)

- 12GB+ RAM:  Most models including Q8_0 (up to 8GB+)

Tip: The app will warn you before loading a model that's too large for your device. Close other apps before loading large models.

Example: A "Llama-3.2-3B-Q4_K_M.gguf" file (2.1GB) needs approximately 4-6GB RAM to run smoothly.