Singapore-based startup Acti launched what it calls the world's first "agentic keyboard" on June 30, a free system-level keyboard for iOS and Android that replaces predictive-text suggestions with an AI agent capable of taking actions directly inside whatever app a user is typing in. Instead of guessing the next word, Acti surfaces an "Acti Bar" and dedicated "Skill Keys" that run pre-built or user-created routines — such as pulling up the weather, checking a calendar, or updating a Notion page — without the user leaving their email, messaging, or social app.

Agents built with plain language, not code

Acti's core differentiator is its "Skills" system: users describe what they want in natural language, and the app generates a reusable routine rather than requiring any coding. The company said early access testers built more than 1,000 Skills in under two weeks ahead of launch, and Skills can be kept private or published to a public marketplace for other users to install. The pitch is that this brings agentic AI — until now mostly confined to browser extensions, coding tools, and enterprise software — into the keyboard layer that touches every app on a phone.

Acti is run by CEO Young Wang, who previously spent roughly a decade at Baidu growing its Facemoji Keyboard product to more than 300 million daily active users, giving the founding team direct experience scaling consumer keyboard software. CTO Mike Sun previously led engineering for Baidu's Yike Album cloud-photo platform. The startup raised a $5.3 million seed round led by BITKRAFT Ventures, which it says will fund engineering and AI hires, deeper on-device intelligence, and growth of the Skills ecosystem.

A privacy-sensitive entry point for agents

Because a keyboard has access to virtually everything a user types, Acti has emphasized a local-first design: personal context is meant to stay on-device by default, and the company says it does not access or store private messages or conversations unless a user explicitly invokes a feature that requires external processing. That framing addresses one of the more obvious objections to an AI agent sitting at the OS input layer — that it would need broad, persistent access to a user's most sensitive text to be useful.

The launch lands as agentic AI products increasingly compete for consumer attention beyond the chatbot interface, following moves by larger players to embed agents into existing surfaces like messaging apps and productivity suites. Acti's bet is that the keyboard, as the one interface present across nearly every app on a phone, is an underexploited distribution point for agentic features — but the company will need to prove that its Skills marketplace and on-device model can scale reliably beyond the early-access cohort that built its first 1,000 routines.