CtxLLM
CtxLLM stores conversation context in a Shortcuts variable or file so LM Mini remembers prior exchanges across runs — synthetic memory without a cloud service.
One-tap install
Build the shortcut below (2 minutes), then share to iCloud from Shortcuts.app to enable one-tap install for others. Update website/data/shortcuts.json with your icloudUrl and re-run this generator.
Siri Phrase
"Hey Siri, continue with context"
Run once after installing, then Siri learns this phrase.
How the shortcut works
- Each run loads prior context from a Shortcuts file or Data jar.
- Your new message is appended and sent to LM Mini.
- LM Mini responds with full thread awareness.
- Updated context is saved back for the next run.
Key features
- Persistent memory via Shortcuts file storage
- Fully local — memory file stays on device
- Works with Ask LM Mini and Voice Chat
- Advanced: combine with LM Mini conversation IDs via URL scheme
Recommended models
Qwen 3 4B with 8k+ context via LM Studio if needed
Connect to LM Studio 32k context model for long memory threads
Tune temperature (0.7) and top-p (0.9) in LM Mini → Settings → Model Parameters. See the full on-device model catalog.
Build it in 60 seconds
- Create a text file in iCloud Drive or Shortcuts folder for context storage.
- Get Contents of File → Combine with new Dictated Text → Ask LM Mini.
- Combine response with thread → Save File.
- Use lmmini://shortcut/voice?conversationId= for in-app persistent threads.
URL scheme (power users)
Add an Open URLs action in Shortcuts.app:
lmmini://shortcut/ask?prompt=Continue%20our%20conversation&onDevice=1Need LM Mini? Free on iOS and Android — on-device GGUF/MLX, LM Studio Connect, voice, and more.
Download LM Mini LM Mini vs Private LLM