Private AI assistant for macOS
Local AI. Web context. No cloud.
SilicIA is a native macOS application powered by local large language models and Apple Silicon. Chat with an on-device AI, search the web and get synthesised summaries with cited sources, or ask questions about your PDFs and images โ all processed locally, with no data ever sent to an external server.
AI inference
100% on-device (local LLM)
Web search
DuckDuckGo + AI synthesis
Document analysis
PDF & images, page citations
Storage
Local chat history (SwiftData)
Features
- Local LLM chat โ multi-turn conversations processed entirely on your Mac, no internet required.
- Web intelligence โ live DuckDuckGo search with AI-synthesised summaries and cited source links.
- Document analysis โ drag-and-drop PDFs or images; get answers with page-precise citations.
- Combined context โ blend web results and documents in a single conversation.
- Privacy by design โ no account, no telemetry, no cloud inference. Your data stays on your device.
- Multi-language โ English and French interfaces.
- Persistent history โ conversations stored locally with SwiftData.
How it works
- Chat: your prompt is sent to a local LLM โ no network call required.
- Search: query DuckDuckGo, scrape top results, feed them as context to the local model.
- Documents: PDF or image content is chunked, ranked, and injected as RAG context.
- Answer: the model reasons over the combined context and responds with cited sources.
Technical highlights
- SwiftUI โ native macOS interface, optimised for Apple Silicon.
- Local LLM โ on-device inference via Apple's Foundation Models framework.
- RAG pipeline โ chunking, ranking, and retrieval-augmented generation.
- PDFKit โ PDF parsing and page-level citation extraction.
- SwiftData โ persistent local chat and PDF conversation history.
Requirements & build
- Requirements: macOS 26+, Apple Silicon.
- Clone:
git clone https://github.com/Eddy-Barraud/SilicIA.git - Open:
open SilicIA.xcodeproj - Build: Xcode
⌘B(or⌘Rto run).