|
|
|
|
|
|
|
|
|
|
What I’m exploring · Local models, context & agent memory
I’m exploring how local models and agent systems can fit into the tools we already use.
- Agent memory & handoff. Keeping project context useful across sessions and tools with ShareMemory.
- On-device inference. Working with WhisperKit, Core ML, and Apple Silicon in NoType.
- Context-aware assistance. Bringing screen and browser context into Wisp when the user asks.
- Hardware-aware engineering. Thinking through latency, memory, and compute budgets alongside model quality.
How I build · A few principles behind the work
- Keep data close. Prefer on-device processing where it makes sense, and make network dependencies explicit.
- Keep the user in control. Context should be intentional; actions should be understandable.
- Make handoffs explicit. Give agents useful state, clear boundaries, and outcomes that can be checked.
- Sweat the small interactions. A shortcut, a preview, or one less window can make a tool worth keeping.
