An AI Agent Now Lives in Your Keyboard

Mid-conversation, you type "tell them I'll be ten minutes late and suggest 7 p.m. instead." Normally what follows is mechanical: you leave the app, open an assistant, copy its reply, come back to paste it, fix the time. This time, none of that. The keyboard read the request, wrote the message, adjusted the hour and sent it, without your thumb ever leaving the chat screen.

That tiny scene sums up the bet made by Acti, a Singapore startup that closed a $5.3 million seed round in late June 2026, led by the fund BITKRAFT. The idea fits in a sentence: the keyboard, that rectangle of keys you reopen hundreds of times a day, is the most obvious place to put an agent that no longer only writes, but acts.

The Keyboard as an Execution Layer

A keyboard, on a phone, is not one app among others. It is the mandatory checkpoint: everything you send out into the world, a message, a search, a password, an address, passes through it. Acti starts from that fact and turns the surface into what its founders call an execution layer, a place from which you trigger actions without switching context.

In practice, the user creates "Skills" by describing, in plain language, what they want: "translate this paragraph into polite German," "turn these notes into a summary," "answer this email in my usual tone." No coding required. Before launch, testers had already built more than a thousand of them in under two weeks. Available on iOS and Android, the tool leans on Google's Gemini models to understand and write.

Behind the product lies a wider ambition: to make the keyboard a personal "context layer," a place that knows your phrasings, your contacts, your habits, and uses them to act quickly. It is the same move running through the whole sector in mid-2026: agents are leaving the chat window we used to visit and slipping into everyday gestures, as close as possible to the moment intent is born.

Where a classic assistant asks you to come to it, in its app or its tab, Acti does the reverse: it descends to where you already are, under your fingers, at the instant you write. The gesture is no longer "open the AI, phrase it, retrieve, paste back," but "type the intent, let it run."

The Shortcut and the Time It Returns

What this shift removes is a friction gone invisible through habit: the round trip. We spend our days copying on one side to paste on the other, stepping out of a conversation to query a model and then stepping back in. Each back-and-forth costs only seconds, but there are dozens a day, and each one breaks the thread of what we were doing.

By housing the agent in the keyboard, Acti compresses that chain into a single move. Intent and execution meet in the same place. For anyone who writes a lot, across several languages, with repetitive tasks, the gain is not trivial: it owes less to raw speed than to continuity, to no longer dropping one task to go and perform another.

Take the traveler who, abroad, has to answer a hotel clerk in a language they barely handle. Yesterday they juggled the messaging app and a translator; today they write their intent in English and the keyboard delivers, inside the conversation itself, a courteous reply in Italian or Japanese. The service is not new, but the place where it is rendered is: there is no longer a seam between wanting and doing.

There is also a form of democratization. Describing a Skill in plain English, without a line of code, puts a small personal automaton into anyone's hands. Whoever answers the same questions each day, fills in the same forms, rewrites the same emails, can cut a tool to their own measure in a sentence. This is the promised autonomy: less dependence on a single service, more gestures handed to a mechanism you tuned yourself.

A Keyboard That Reads Everything

Yet this convenience settles into the most sensitive spot on the phone. A keyboard sees, by nature, absolutely everything: your intimate messages, your logins, your banking codes, your searches no one should read. Handing that vantage point to an autonomous agent means opening it the widest possible window onto your digital life.

Acti answers the objection with a "local-first" principle: your personal context would stay, by default, on the device. The promise is serious, but it runs into a nuance. Understanding and writing your requests rely on the Gemini models, which run, for their part, on Google's servers. "Local-first" protects the memory of your habits; it does not mean that every sentence handed to the agent stays on the phone. Between what the keyboard keeps and what it sends off to be processed, the line is worth reading closely.

Then there is the matter of permissions. On iOS as on Android, a third-party keyboard that truly acts demands "full access," that is, the ability to see what you type. That access is the price of the convenience, and it is not negotiated piecemeal. Finally comes dependence: the more useful the agent grows, the more gestures we entrust to it, and the costlier it becomes to do without it, or to find out one day that the company holding it has changed hands, price or rules.

The Most Contested Surface on the Phone

For a long time the keyboard served only one purpose, writing. Making it the seat of an agent that acts is to admit it has become the most direct door to our intentions. What is at stake there is bigger than Acti and its five million: the giants that already own our keyboards, Google and Apple foremost, are watching the same ground.

The market to win, then, is not that of the smartest assistant, but of the place where you meet it. The convenience is real, measurable in the round trips you no longer make. The reckoning is colder: agreeing that an agent should read everything you type, in exchange for the gestures it spares you, comes down to deciding what you would rather keep to yourself. The shortcut is appealing as long as you know, precisely, what it passes through to go faster.