Acti puts AI agents directly into your smartphone keyboard

The keyboard framing is less about UX novelty and more about bypassing the app store and OS permission layers that have bottlenecked AI assistant distribution. Acti is essentially betting that the input method editor (IME) slot, which users rarely revoke once granted, is a more durable foothold than any standalone app.

This sits in direct tension with the infrastructure story we covered the same day: Anthropic's Claude Sonnet 5 launch as a cheaper agent runtime. If inference costs keep falling, the bottleneck for agent adoption shifts from model expense to distribution and invocation surface, which is precisely the gap Acti is trying to occupy. A cheaper Sonnet 5 actually makes Acti's proposition more viable, since the economics of running LLM calls on every keyboard shortcut improve as per-token costs drop. The open question is whether Acti licenses model access or lets users bring their own keys, because that choice determines whether they become a thin UI layer or a meaningful aggregator.

Watch whether any of the major keyboard incumbents (Gboard, SwiftKey) announce agent integration within the next two quarters. If they do, Acti's window as a differentiated product closes fast and the company becomes an acquisition target rather than a standalone platform.