BAK! — Support

Welcome to the BAK! support page. Below you'll find answers to common questions, troubleshooting tips, and ways to reach us for further assistance.

First Launch

When you open BAK! for the first time, you'll need to download or import a GGUF model before you can start chatting. You can:

• Browse Hugging Face: Tap the model picker and use the built-in Hugging Face browser to search, filter, and download GGUF models directly
• Import a local file: If you already have a GGUF file, use the "Load Custom Model" option to import it
• Download from URL: Paste a direct download link to a GGUF file

Recommended Models by Device

App Crashes on Model Load

This usually means the model is too large for your device's available memory. Try:

• Using a smaller quantization (Q4_K_M instead of Q8_0)
• Using a smaller model (0.8B or 2B instead of 4B or 9B)
• Closing other apps to free up memory
• Reducing the context length in Settings

Slow Generation Speed

Generation speed depends on your device and model size. To improve it:

• Increase GPU layers (ngl) in Settings — set to 999 to offload all layers to GPU
• Use a smaller model or lower quantization
• Reduce the context length

Model Download Fails

If a download from Hugging Face fails:

• Check your internet connection
• Ensure you have enough free storage space
• Try downloading over Wi-Fi instead of cellular
• Retry the download — temporary network issues are common for large files

RAG Not Working

For RAG (Retrieval-Augmented Generation) to work, you need:

• An embedding model loaded (e.g., bge-m3)
• At least one document uploaded to the document store
• RAG enabled in Settings

Chat

BAK! provides a full chat interface with Markdown rendering, code blocks, and collapsible thinking blocks. You can edit and regenerate any message.

RAG & Documents

Upload documents (PDF, TXT) and BAK! will chunk, embed, and search them to provide relevant context to the LLM. Requires an embedding model.

Tool Calling

Enable tool calling in Settings to let the model use web search, fetch URLs, search documents, and interact with MCP servers autonomously.

Avatar Mode

Create a talking character with chroma-key video, persona documents, and automatic TTS. Great for interactive experiences.

API Server

Expose your loaded models as an OpenAI-compatible API on your local network, accessible by other apps and devices.

Remote APIs

Connect to OpenRouter, OpenAI, or any OpenAI-compatible endpoint for cloud-based inference. Mix local and remote models per task.

Q: Does BAK! require an internet connection?

A: No. With local models, BAK! works completely offline. Internet is only needed for optional features like model downloads, web search, remote APIs, and TTS.

Q: Where are my conversations stored?

A: All conversations are stored locally on your device in the app's Documents/chats/ directory as JSON files. They are never uploaded anywhere.

Q: Can I use my own API keys?

A: Yes. BAK! supports any OpenAI-compatible API endpoint. You provide your own API key and base URL in Settings.

Q: What model formats are supported?

A: BAK! supports GGUF format models, which is the standard format used by llama.cpp.

Q: Is my data private?

A: Yes. BAK! collects no data whatsoever. Everything stays on your device. See our Privacy Policy for details.

Q: Does BAK! support Turkish?

A: Yes! The entire UI is available in both English and Turkish. The language follows your device settings.

If you can't find an answer to your question above, we're happy to help:

• Email: [email protected]
• GitHub: github.com/salihkes
• Response Time: We aim to respond within 48 hours

When reporting a bug, please include:

• Your device model (e.g., iPad Pro M2)
• iOS version
• The model you were using
• Steps to reproduce the issue