Acti is trying to move AI out of the chat window and into the place where people already spend much of their time typing. With a new agentic keyboard for iOS and Android, the Singapore-based startup says users can do more than accept predictive text or launch a chatbot. They can trigger actions inside other apps through programmable “Skills” — keyboard shortcuts that carry out multi-step tasks across messaging, email, social media, and beyond.
That shift matters because it reframes the keyboard as an execution surface, not just an input method. In Acti’s version of the mobile stack, the keyboard becomes a cross-app control plane: the user types intent once, and the system coordinates the rest. For technical readers, the interesting part is not the headline claim that the keyboard is “agentic,” but the operational problem it exposes. The harder work in mobile AI is no longer text generation. It is orchestration across fragmented app contexts, with the right permissions, low enough latency, and predictable behavior when the underlying apps were never designed to be automated together.
Acti says the action layer is powered by Google’s Gemini models. That choice is telling. In a keyboard product, the model is not just generating language; it is being used to interpret intent, map that intent onto a task plan, and drive the steps needed to complete the task inside native apps. Founder and CEO Young Wang frames the company’s bet as a user-owned context layer that spans apps rather than living inside a single platform. In his view, the limitation of many current AI agents is that context remains trapped inside silos. Acti, by sitting across them, is trying to assemble that missing state into something the agent can use.
That architecture suggests a layered system rather than a single prompt-to-action pipeline. A user action, such as asking the keyboard to translate a message before sending it, likely has to be resolved into a sequence: identify the source content, determine the target app or field, call the model to interpret the request, and then execute the steps in the destination app. The value proposition is obvious. The technical burden is less so, because each additional hop increases room for failure. Context can go stale. UI states can change. Permissions may not be available. And if the system needs to hand off between apps on the fly, latency quickly becomes visible to the user.
That is why the most interesting part of Acti’s launch is not the existence of Skills, but what Skills imply about how mobile agents must behave to be useful. A shortcut that triggers a multi-step task is only compelling if it feels faster than doing the task manually, or at least reduces enough friction to justify the pause. In a keyboard, that bar is unforgiving. People notice delays immediately, especially in messaging and other high-frequency workflows where the cadence of typing matters. If the orchestration layer has to wait on model inference, app switching, and permission checks, the experience can collapse into something that feels clever in demos but clumsy in practice.
The same is true for reliability. A keyboard sits at the boundary between user intent and action, which means failure modes are more visible than they would be in a standalone assistant. If a Skill misreads the request, targets the wrong app, or completes only part of a multi-step operation, the user may not only lose time but also lose trust in the entire system. That makes consistency as important as capability. The more Acti leans into cross-app action enablement, the more its product becomes a test of whether agentic interfaces can be dependable enough for routine use, not just occasional automation.
Privacy and consent are the other constraints that now define the category. Acti’s pitch hinges on a context layer that belongs to the user, but putting that context into action means the system may have to inspect content across apps, infer intent from messages, and move data between services. For users, that raises obvious questions about what is processed on-device, what is sent to the model, what is stored, and what approvals are required before a Skill can act. For enterprise buyers and regulated environments, those questions are not peripheral. They determine whether the keyboard can be adopted at all.
Platform policy is another boundary Acti will have to navigate carefully. iOS and Android both allow rich input methods, but the moment a keyboard starts executing actions across apps, it moves closer to the area where operating system rules, app permissions, and store policies all matter. Even if the mechanics are technically possible, the product still has to survive the practical realities of mobile platforms that were not built around a universal automation layer. That means Acti’s rollout is as much a policy exercise as a product one.
This is also where the broader competitive frame gets more interesting. Many AI interfaces still live inside a single app, where the model can answer questions or draft text but does not need to coordinate the user’s other software. Acti is pushing in a different direction: an interface that starts at the keyboard and claims a shared context across apps. If that model works, it would be less like a chatbot and more like a mobile operating layer for intent. If it does not, the bottleneck will probably not be model quality alone. It will be the friction imposed by permissions, platform constraints, and the sheer difficulty of making cross-app automation feel immediate.
The startup’s launch, covered by TechCrunch, arrives at a moment when the category is moving from novelty toward implementation. That matters because the success criteria are now concrete. Readers should watch whether Skills remain a small set of polished shortcuts or evolve into a broader developer-facing automation surface. They should watch how often users actually invoke them, how much latency the product can hide, and whether the system can handle common tasks without breaking the flow of typing. They should also watch whether Acti exposes enough tooling for developers to build Skills that are safe, understandable, and maintainable rather than brittle one-off automations.
The larger question is whether the keyboard can become a durable host for AI agents at all. Acti’s launch makes a strong case that the answer could be yes, at least for certain workflows. But the category will be won not by the broadest promise of autonomy, but by the narrow mechanics of orchestration: context management, permission design, failure handling, and latency discipline. That is a harder story to tell than “AI in your keyboard,” but it is also the one that will determine whether the product becomes infrastructure or just another clever layer on top of mobile apps.
