Skip to content

v0.9.0-beta.4

Date: 2026-04-14 Download: GitHub Release

Ask AI now remembers what you discussed. Follow-up questions work naturally — ask “What are the ingredients in carbonara?” then “How long should I cook it?” and the AI knows “it” means carbonara. Start a New Chat to reset context.

Rate limit errors now explain what happened in plain English and suggest how to fix it, instead of showing raw HTTP codes.

Golden Retriever indexes your local files (documents, images, videos, audio) and makes them searchable using AI. Ask questions in natural language and get answers with source citations.

  • macOS 14.0 or later
  • A Gemini API key (free at aistudio.google.com/apikey)
  • On first launch, click “Always Allow” for each keychain prompt (3 prompts, one-time only)
  • 3 folders maximum
  • 50 files maximum
  • Supported file types: PDF, Image, Text, CSV, JSON, XML, Video, Audio
  • Office formats (DOCX, PPTX, XLSX), iWork (Pages, Keynote, Numbers), RTF, and HTML will be available in a future paid tier
  • Embedding is throttled to stay within API limits — initial indexing takes longer
  • 20 requests/day limit for Gemini Flash (used for search/QA) — heavy use may return rate limit errors
  • If you hit rate limits, wait an hour and try again
  • Consider upgrading to Gemini paid tier in Google AI Studio for higher limits
  • No billing/subscription management (beta is free)
  • No Vertex AI / Google Cloud option (Gemini API only)
  • No data residency controls
  • No database encryption
  • 3 keychain prompts on first launch of notarized build
  • No way to resume a cancelled pipeline — must re-run from scratch
  • 30-day trial from first launch
  • Hard deadline: June 11, 2026
  • App shows expiry screen after trial ends
  • Anonymous crash/usage reports are opt-in during setup
  • No document content, search queries, or file names are ever sent
  • Can be disabled anytime in Settings > Privacy
  • See the in-app Legal Agreement for full details

Report bugs and feedback via GitHub Issues or directly to the team.