Privacy Policy for Bastion Chat
Overview
Bastion Chat is designed with privacy as our core principle. This Privacy Policy explains how we handle information in our AI chat assistant application.
Our Privacy Commitment
We do not collect, store, transmit, or have access to any of your personal data, conversations, or documents.
Information We Do NOT Collect
Your conversations with AI models remain entirely on your device
Uploaded PDFs, text files, and web content stay local
Speech recordings and transcriptions are processed locally
No names, email addresses, or contact information collected
No telemetry, crash reports, or usage statistics transmitted
No device IDs, IP addresses, or system information collected
No location tracking or geo-data collection
How Bastion Chat Works
🔒 100% Local Processing
- All AI model inference happens on your device
- Conversations, document processing, and voice recognition occur locally
- No data is sent to external servers for processing
📱 Local Storage Only
- Chat history is stored in a local SQLite database on your device
- Documents are processed and stored locally using vector embeddings
- All user preferences and settings remain on your device
🌐 Limited Network Usage
The app only connects to the internet for:
- Model Downloads: Downloading AI models from Hugging Face repositories
- Web Document Ingestion: When you explicitly choose to add a web URL to your knowledge base
- Web Search (optional, off by default): When you turn on Web access, the AI can send a search query to DuckDuckGo and open the web pages it finds to answer your question
- MCP Servers (optional): Only servers you add yourself in Settings, and only the requests you allow the AI to make to them
Network requests are made directly to:
- Hugging Face (huggingface.co): For downloading AI models
- User-specified websites: Only when you provide a URL to add to your knowledge base
- DuckDuckGo (duckduckgo.com) and the result pages: Only while Web access is turned on
- MCP servers you configure: Only if you add one
When the device is offline, web tools are not offered to the AI at all. Your documents and conversation history are never uploaded; a web search sends only the search query the AI writes.
No data about you or your usage is transmitted during these connections.
Permissions We Request
Data Security
🔐 Device-Level Security
- Data is protected by your device's built-in security (encryption at rest)
- Lock BastionChat with Face ID or Touch ID using iOS's built-in app protection (touch and hold the app icon, then choose Require Face ID)
- All processing happens within the app's sandboxed environment
🛡️ No Cloud Vulnerabilities
- No cloud storage means no risk of remote data breaches
- No external API keys or tokens stored
- No third-party analytics or tracking services
Your Rights
Since we don't collect your data:
- No Data to Delete: All data is yours and remains on your device
- No Data to Export: Your data never leaves your device unless you choose to back it up
- Complete Control: You can delete the app to remove all associated data
Children's Privacy
Bastion Chat does not collect personal information from anyone, including children under 13. The app's privacy-first design ensures no data collection regardless of user age.
Changes to Privacy Policy
We will update this policy if our privacy practices change. Since we don't collect contact information, updates will be communicated through app updates.
Contact Information
For privacy-related questions:
- GitHub Issues: Report concerns through our public repository
- Documentation: Refer to our technical documentation for architecture details
Compliance
This privacy policy is designed to comply with:
- Apple App Store Guidelines
- General Data Protection Regulation (GDPR)
- California Consumer Privacy Act (CCPA)
- Children's Online Privacy Protection Act (COPPA)
Since we don't collect personal data, compliance is achieved through our privacy-by-design architecture.
Summary: Bastion Chat is built on the principle that your conversations, documents, and personal data should remain private and under your control. Our local-first architecture ensures your privacy is protected by design, not just by policy.
Your AI, Your Data, Your Device.
Acknowledgments & Third-Party Licenses
Bastion Chat makes use of several open-source libraries and AI models. We gratefully acknowledge the contributions of these projects and their maintainers.
Swift Package Dependencies
NetworkImage
Repository: https://github.com/gonzalezreal/NetworkImage
License: MIT License
Purpose: Efficient image loading and caching
Author: Guille Gonzalez
SwiftSoup
Repository: https://github.com/scinfu/SwiftSoup
License: MIT License
Purpose: HTML parsing for web content ingestion
Author: Nabil Chatbi
swift-markdown-ui
Repository: https://github.com/gonzalezreal/swift-markdown-ui
License: MIT License
Purpose: Markdown rendering in SwiftUI
Author: Guille Gonzalez
swift-atomics
Repository: https://github.com/apple/swift-atomics.git
License: Apache License 2.0
Purpose: Atomic operations for thread-safe programming
Author: Apple Inc.
swift-cmark
Repository: https://github.com/swiftlang/swift-cmark
License: BSD 2-Clause License (cmark), see repository for bundled components
Purpose: CommonMark parsing
Author: John MacFarlane and the Swift Project Authors
MCP Swift SDK
Repository: https://github.com/modelcontextprotocol/swift-sdk
License: Apache License 2.0 (some earlier contributions remain under the MIT License)
Purpose: Model Context Protocol client for optional external tools
Author: Model Context Protocol contributors
EventSource
Repository: https://github.com/mattt/eventsource
License: MIT License
Purpose: Server-sent events for MCP connections
Author: Mattt
swift-nio
Repository: https://github.com/apple/swift-nio
License: Apache License 2.0
Purpose: Networking primitives used by the MCP SDK
Author: Apple Inc.
swift-log, swift-collections, swift-system
Repositories: swift-log, swift-collections, swift-system
License: Apache License 2.0
Purpose: Logging, data structures and system interfaces
Author: Apple Inc.
Core Inference Engine
llama.cpp
Repository: https://github.com/ggml-org/llama.cpp
License: MIT License
Purpose: High-performance LLM inference engine
Author: Georgi Gerganov and contributors
Note: Foundation for GGUF model inference and optimization
AI Models & Licensing
Important: The following AI models are available for download through Bastion Chat. Users are responsible for compliance with individual model licenses.
Qwen Models
Models: Qwen2.5-0.5B-Instruct, Qwen3-1.7B, Qwen3-4B, Qwen3.5-2B, Qwen3.5-4B, Qwen3.5-9B, Qwen3.6-27B
License: Apache License 2.0
Provider: Alibaba Cloud (Qwen team)
License URL: View License
Gemma 4 Models
Models: Gemma 4 E2B, E4B, 12B, 26B-A4B and 31B (instruction-tuned)
License: Apache License 2.0
Provider: Google LLC
License URL: View License
Gemma 3 Models
Models: Gemma 3 270M, 1B and 4B (instruction-tuned)
License: Gemma Terms of Use
Provider: Google LLC
License URL: View License
Llama Models
Models: Llama-3.2-1B-Instruct, Llama-3.2-3B-Instruct
License: Llama 3.2 Community License Agreement
Provider: Meta Platforms, Inc.
License URL: View License
Hermes Models
Models: Hermes-3-Llama-3.2-3B, DeepHermes-3-Llama-3-3B-Preview
License: Llama 3 Community License (fine-tunes of Llama 3.2, subject to the base model license)
Provider: Nous Research
License URL: View License
Phi Models
Models: Phi-4-mini-instruct
License: MIT License
Provider: Microsoft Corporation
License URL: View License
LLaVA-Phi-3-mini
Models: llava-phi-3-mini (vision)
License: No license declared on the model card — see the repository before use
Provider: XTuner
License URL: View License
LFM Models
Models: LFM2-1.2B, LFM2.5-1.2B-Instruct
License: LFM Open License v1.0
Provider: Liquid AI
License URL: View License
SmolLM3
Models: SmolLM3-3B
License: Apache License 2.0
Provider: Hugging Face
License URL: View License
Jan
Models: Jan-v1-4B
License: Apache License 2.0
Provider: Menlo Research (Jan)
License URL: View License
Lucy
Models: Lucy-128k
License: Apache License 2.0 (based on Qwen3-1.7B)
Provider: Menlo Research
License URL: View License
Bonsai
Models: Bonsai-27B (experimental)
License: Apache License 2.0 (based on Qwen3.6-27B)
Provider: Prism ML
License URL: View License
Licenses as declared on each model's Hugging Face page, checked October 13, 2026. Models you import yourself (any GGUF file) are subject to their own licenses.
BERT Model for Embeddings
all-MiniLM-L6-v2
License: Apache License 2.0
Provider: Microsoft Corporation / Hugging Face
Repository: sentence-transformers/all-MiniLM-L6-v2
Purpose: Semantic embeddings for document search
Usage: Local embedding generation for RAG functionality
License URL: View License
Model Usage Terms
Important Notes for Users:
- Model Downloads: Models are downloaded directly from their respective repositories
- License Compliance: Users must comply with individual model licenses
- Commercial Use: Check specific model licenses for commercial usage terms
- Attribution: Some models may require attribution in commercial applications
- Redistribution: Bastion Chat does not redistribute models; they are downloaded on-demand
The user is solely responsible for ensuring compliance with all applicable licenses when using AI models through Bastion Chat. The app facilitates access to these models but does not grant any additional rights beyond those specified in the original model licenses.
Additional Technologies
Apple Frameworks
Frameworks: SwiftUI, Foundation, Combine, AVFoundation, Speech, CoreML
License: Apple Software License Agreement
Provider: Apple Inc.
Purpose: iOS app development and system integration
Last Updated: October 13, 2026
Bastion Chat Version: 1.1.0
We are grateful to all open-source contributors who make projects like Bastion Chat possible.