The private alternative to Perplexity.
Search the web. Get answers on-device.
A native iOS 26 / macOS 26 answer engine that combines web search with on-device AI. AISight searches the web via a privacy-respecting SearXNG instance (default provided, or self-host your own) and synthesizes answers using Apple's FoundationModels framework (Apple Intelligence) — all AI processing happens privately on your device. Zero external dependencies.
Every major answer engine — Perplexity, Google AI Overviews, Bing Copilot — sends your queries to cloud AI services, building a profile of what you search. AISight takes a different approach: the AI runs entirely on your device via Apple Intelligence, so your questions never leave your phone or Mac. You control the search backend too (self-hosted SearXNG), meaning no single company sees both what you search and what you ask about it. It's the answer engine for people who think privacy shouldn't be a premium feature.
- Answer factual, encyclopedic, and "how-to" questions with cited sources
- Deep Search — multiple AI research passes for complex questions (Pro)
- Search the web via SearXNG (aggregates Google, Bing, Brave, and more)
- On-device AI synthesis — your data is never sent to a cloud AI service
- Stream responses token by token with inline citations
- Reciprocal Rank Fusion (RRF) for multi-engine result ranking
- Persist query history locally via SwiftData
- Universal app — runs natively on iPhone, iPad, and Mac
AISight Pro ($4.99, one-time) unlocks unlimited searches, Deep Search, and custom SearXNG server.
- Multi-hop reasoning beyond Deep Search (on-device model is ~3B parameters)
- Real-time news with high freshness guarantees
- Complex math or coding assistance
- Image understanding (text-only)
- Work without internet (search requires connectivity)
- Run on devices older than iPhone 15 Pro / iOS 26
- Xcode 26+
- iOS 26.0 / macOS 26.0 deployment target
- Apple Intelligence enabled on device
- A SearXNG instance (local Docker or remote)
A Docker Compose setup is included for local development:
cd searxng
docker compose up -dThis starts SearXNG on http://localhost:8888 with Google, Bing, Brave, and Wikipedia enabled.
For production, deploy your own instance following the SearXNG Docker guide.
- Open
AISight/AISight.xcodeprojin Xcode 26 - The app ships with a default SearXNG instance — for local development, change the URL to
http://localhost:8888in the app's Settings tab or inAISight/App/AppConfig.swift - Select your target (iPhone simulator or My Mac) and Cmd+R
macOS note: Add the "Outgoing Connections (Client)" capability in Signing & Capabilities for network access.
Physical device note: Replace
localhostwith your Mac's local IP in the app's Settings tab.
User Query → SearXNG (search) → ContentFetcher (HTML→text) → FoundationModels (on-device AI) → Streamed Answer with Citations → SwiftData (history)
| Layer | Path | Responsibility |
|---|---|---|
| App | AISight/App/ |
Entry point, config, global state |
| Core/AI | AISight/Core/AI/ |
FoundationModels session, Deep Search pipeline |
| Core/Search | AISight/Core/Search/ |
SearXNG API client, RRF ranking, models |
| Core/Fetching | AISight/Core/Fetching/ |
URL → clean text extraction |
| Core/Persistence | AISight/Core/Persistence/ |
SwiftData models and store |
| Features | AISight/Features/ |
Search, History, Onboarding, Settings |
| UI/Components | AISight/UI/Components/ |
CitationText, SourceCard, etc. |
Tech stack: Swift 6, SwiftUI, FoundationModels, SwiftData, URLSession. No external packages.
All AI inference runs on-device via Apple Intelligence. Data that leaves your device:
- Search queries sent to a SearXNG instance
- HTTP requests to source URLs for content fetching
No analytics. No tracking. No third-party AI services.
Contributions welcome. Open an issue before submitting large PRs.
Built by a solo developer in Spain. Entire infrastructure costs under €5/month.