SwiftZilla

Discussion (1 comments)

Showing 1 comments

vituu

19 days ago

I've been getting increasingly frustrated with how much current LLMs (Claude, GPT, Gemini, etc.) hallucinate on modern Swift, especially since Swift 6.0 and now with 6.1/6.2 changes rolling out. They confidently suggest deprecated SwiftUI modifiers, wrong concurrency patterns (actors, Sendable violations, etc.), or APIs that changed in recent betas.

So I built a narrow, focused RAG just for Apple/Swift development.

What it covers: - Full official Apple documentation (~100k+ pages) - Swift API Design Guidelines - All WWDC session transcripts - Swift Evolution proposals (SEPs)

It's deliberately not a general-purpose RAG. Only Apple ecosystem stuff, kept fresh.

It speaks MCP (Model Context Protocol) natively — SSE or STDIO — so it plugs straight into Cursor, Windsurf, Claude Desktop, or any custom agent that supports MCP.

I've been using it myself and shared it with a few iOS/macOS/tvOS friends over this month. Pricing is $3/month (basically covers hosting/indexing costs; cancel anytime). There's a free tier with limits, and right now a shared community pool of ~9–10M tokens is open (bypasses daily limits while it lasts).

https://swiftzilla.dev

Curious to hear from others who use AI agents for Swift work: How do you currently deal with hallucinations on Swift 6+ features (Observation, macros, strict concurrency, etc.)? Do you just manually paste docs, use web search in agents, or something else?

Open to any feedback, especially if you've tried similar setups and run into the same issues.