BAK!
Büyük Dil Modeli Araç Kutusu
Your on-device LLM toolkit for iOS & iPadOS
Run powerful AI models directly on your iPhone and iPad — completely offline, fully private, with no compromises.
Everything you need, on your device
BAK! brings the full power of large language models to your Apple devices — no cloud required.
Private by Design
All AI processing happens on your device. Zero data collection, zero telemetry, zero cloud dependency. Your conversations never leave your device.
On-Device Inference
Powered by llama.cpp, BAK! runs GGUF models with full GPU acceleration using Apple's Metal framework. No internet required.
Remote APIs
Connect to OpenRouter, OpenAI, or any OpenAI-compatible endpoint when you want cloud-powered models. Mix local and remote per conversation.
RAG & Documents
Upload PDFs and text files. BAK! chunks, embeds, and searches them to give your LLM relevant context — all processed locally.
Tool Calling & MCP
Let models autonomously browse the web, fetch URLs, search documents, and interact with MCP servers using a ReAct loop.
Vision Models
Send images to multimodal models for description, analysis, and visual Q&A — both with local and remote vision models.
Avatar Mode
Create a talking AI character with chroma-key video, persona documents, and automatic text-to-speech for interactive experiences.
API Server
Expose your loaded models as an OpenAI-compatible API on your local network. Other apps and devices can connect to your iPad or iPhone.
See BAK! in action
A clean, native interface designed for iPad's large screen and iPhone's portability.
Truly Local AI
BAK! is built on llama.cpp, the industry-standard inference engine for running large language models on consumer hardware. With full Metal GPU acceleration, your iPad or iPhone becomes a powerful AI workstation. Download models directly from Hugging Face, import GGUF files, or paste download URLs — BAK! handles the rest.
Autonomous Tool Use
Enable tool calling and watch your AI come alive. BAK! implements a ReAct agent loop where the model can decide to search the web, fetch live data from URLs, query your documents, or interact with MCP (Model Context Protocol) servers — all autonomously within a single conversation turn.
Your Personal Knowledge Base
Upload PDFs and text documents, and BAK! will chunk them, generate embeddings with models like BGE-M3, and store them locally. When you ask a question, relevant passages are automatically retrieved and injected into context. Combined with a reranker model, BAK! delivers precise, document-grounded answers.
Zero Data Collection
BAK! contains no analytics SDKs, no advertising frameworks, no tracking mechanisms, and no telemetry of any kind. We don't collect personal information, usage statistics, device identifiers, location data, or crash reports. Everything stays on your device and is removed when you delete the app.
Recommended Models by Device
| Device | Model | Size |
|---|---|---|
| iPhone 12 mini (4 GB) | Qwen3.5-0.8B Q4_K_M | 533 MB |
| iPhone 14/15 (6 GB) | Qwen3.5-2B Q4_K_M | 1.28 GB |
| iPad M1+ (8 GB) | Qwen3.5-4B Q4_K_M | 2.74 GB |
| iPad M1+ (16 GB) | Qwen3.5-9B Q4_K_M | 5.68 GB |
Get Started
BAK! is currently in beta testing via Apple TestFlight.