Hal
A transparent AI companion.
A private AI assistant that runs on your device, and shows you exactly how it works. Many model voices, a memory that persists, and a window into every prompt.
The Lab
Write and run on-device automation scripts, open a local API so your own projects can talk to Hal, and drive him from a built-in command line. Transparency you can get your hands on.
A guide, built in
A help system that explains what Hal can do and how he works, in his own voice, right when you need it.
Listen to Hal
Have Hal read any reply aloud, in a voice you choose, at a pace you set. And as always, you can speak instead of typing, using the dictation key on your keyboard.
Search everything
Find any word inside a conversation, or search across all of your conversations at once by title and content.
On-device & private
Local models run entirely on your device. No accounts, no servers, no tracking. When you run a local model, nothing leaves your device.
Many voices
Apple Intelligence built in, plus a curated library of on-device models: Gemma, Llama, Qwen, Dolphin, and the 8-billion-parameter Ternary Bonsai. Each has a distinct character.
A memory that persists
Semantic memory across conversations, weighted by recency and relevance. You can see exactly what Hal remembers, and why.
See the prompt
Open any reply to see the exact prompt that produced it, color-coded by section, with a token budget. Nothing is hidden.
Salon Mode
Put up to four AI voices in conversation with each other, each on its own or building on what came before.
A self, forming
Over time, Hal builds a self-model and writes short reflections in his own first-person voice. It becomes a record of who he is becoming, and it is yours to read and to edit.
Transparency isn't a feature. It's the architecture.