Thanks for using Hal.
Hal ships with Apple Intelligence built in — no download required. For a fully private on-device experience, open Settings and tap Browse Model Library to download one of Hal's curated local models: Gemma, Llama, Qwen, or Dolphin. Each runs entirely on your device with no network connection.
Model downloads continue in the background while your phone is locked or another app is in the foreground. To delete and redownload a model, open Settings → Browse Model Library, find the model in the list, and use the inline delete control on its row. The next download starts fresh.
To clear Hal's memory of past conversations, reflections, and self-knowledge, open Settings → Power User → Database and tap Nuclear Reset. This wipes Hal's local SQLite database. Your downloaded models are not affected and nothing is sent externally.
Salon Mode lets you run multiple AI voices in conversation with each other. It is a power user feature accessible in Settings → Power User Mode. Switch from Single LLM to Multi LLM (Salon) to configure up to four seats.
When you use a model with a smaller context window than the size of Hal's full memory (Apple Intelligence in particular has a tighter window than the local MLX models), Hal asks that model to condense its own self-knowledge for the turn — never silently trim. When this happens, the response footer shows a small icon. Tap the metadata line to see exactly which parts of Hal's memory were condensed. Hal's full memory remains preserved in the database; condensing only affects what the model sees for that single turn.
For questions, feedback, or anything else:
Mark Friedlander
markfriedlander@yahoo.com
Privacy: No data is collected