How It Works: Acti Bar, Skill Keys, and the Skill Builder

The product has three core layers:

• Acti Bar – Replaces the traditional space bar. The user types a request in plain language (e.g., "create a Google Meet link for 3 PM"), then long-presses the Acti Bar. The AI agent interprets the intent and returns a ready-to-use result (link, reply, summary, or action) that the user can preview and tap to insert into the current app .
• Skill Keys – Customizable key bindings that allow a single tap or hold to instantly trigger a specific workflow, such as "summarize this text" or "check my calendar" .
• Natural-Language Skill Builder – Users can describe a new workflow in plain English (e.g., "Get current weather for {location} and format as a short summary") with no coding required. The system assembles the prompt, tools, and output into a reusable Skill that can be kept private or published to the Skill Hub marketplace .
• 150+ Integrations – Acti supports connections to applications (Gmail, Slack, Notion, Google Calendar, Google Meet, Calendly) via secure OAuth and 150+ APIs (weather, finance, crypto, sports, maps, translation, news, and more) that can be wired into custom Skills without separate setup .

Privacy Questions from Keyboard Access

Because Acti is a system keyboard that can read text fields to perform actions, it raises several privacy concerns:

• Access to all typed content – On iOS, enabling "Allow Full Access" is required for the keyboard to contact Acti's servers; on Android, a system warning about keyboard access appears during setup .
• Acti's stated design protection – Acti's Terms of Service and Privacy Policy state that the keyboard does not maintain a remote keylog and does not collect, store, or transmit typed content to its servers for storage or training unless the user explicitly triggers a Skill that needs that text as input .
• What gets sent, and when – When a Skill is triggered, only the specific input the Skill is configured to read (e.g., clipboard content, selected text, or a runtime input prompt) is transmitted to Acti's services and to third-party AI model providers to fulfill the request .
• Connected tool risk – Users can link accounts (Gmail, Slack, Notion) via OAuth. Acti says it does not silently sync or mine those accounts — it only calls them when a user explicitly triggers a Skill that requires the connection .
• Trust gap – The privacy model relies on the user understanding when "explicit trigger" applies, and on the technical enforcement that the keyboard's network access is scoped only to those moments. The Terms and Privacy Policy are the legal boundary, but independent security audits or code-level verification were not found in the available sources.

Broader AI Industry Trend: Embedding Agents into Mobile Surfaces

Acti's launch is a concrete example of a major 2025–2026 industry shift: embedding AI agents directly into existing user interfaces rather than requiring users to open a separate chatbot app.

• Gartner data – 80% of enterprise applications shipped or updated in Q1 2026 now embed at least one AI agent, up from 33% in 2024 .
• IDC forecast – AI copilots are expected to be embedded in nearly 80% of enterprise workplace applications by 2026 .
• The "agentic leap" – Google Cloud and other analysts describe 2026 as the year AI shifts from answering questions to orchestrating end-to-end workflows semi-autonomously .
• Keyboard as the insertion point – Acti takes this trend to the mobile keyboard specifically, arguing that "ChatGPT and Siri are places you go; Acti is already where you're typing" . This mirrors the broader strategy of turning the lowest-friction surface (the text field) into the agent's entry point, rather than creating a separate app or overlay.

In short: Acti is a Singapore-launched, AI-native keyboard that replaces predictive typing with intent-to-action execution, addresses privacy with a "send only on explicit trigger" design, and fits squarely into the 2026 industry push to embed autonomous agents into the interfaces users already use every day.

How It Works: Acti Bar, Skill Keys, and the Skill Builder

The product has three core layers:

• Acti Bar – Replaces the traditional space bar. The user types a request in plain language (e.g., "create a Google Meet link for 3 PM"), then long-presses the Acti Bar. The AI agent interprets the intent and returns a ready-to-use result (link, reply, summary, or action) that the user can preview and tap to insert into the current app .
• Skill Keys – Customizable key bindings that allow a single tap or hold to instantly trigger a specific workflow, such as "summarize this text" or "check my calendar" .
• Natural-Language Skill Builder – Users can describe a new workflow in plain English (e.g., "Get current weather for {location} and format as a short summary") with no coding required. The system assembles the prompt, tools, and output into a reusable Skill that can be kept private or published to the Skill Hub marketplace .
• 150+ Integrations – Acti supports connections to applications (Gmail, Slack, Notion, Google Calendar, Google Meet, Calendly) via secure OAuth and 150+ APIs (weather, finance, crypto, sports, maps, translation, news, and more) that can be wired into custom Skills without separate setup .

Privacy Questions from Keyboard Access

Because Acti is a system keyboard that can read text fields to perform actions, it raises several privacy concerns:

• Access to all typed content – On iOS, enabling "Allow Full Access" is required for the keyboard to contact Acti's servers; on Android, a system warning about keyboard access appears during setup .
• Acti's stated design protection – Acti's Terms of Service and Privacy Policy state that the keyboard does not maintain a remote keylog and does not collect, store, or transmit typed content to its servers for storage or training unless the user explicitly triggers a Skill that needs that text as input .
• What gets sent, and when – When a Skill is triggered, only the specific input the Skill is configured to read (e.g., clipboard content, selected text, or a runtime input prompt) is transmitted to Acti's services and to third-party AI model providers to fulfill the request .
• Connected tool risk – Users can link accounts (Gmail, Slack, Notion) via OAuth. Acti says it does not silently sync or mine those accounts — it only calls them when a user explicitly triggers a Skill that requires the connection .
• Trust gap – The privacy model relies on the user understanding when "explicit trigger" applies, and on the technical enforcement that the keyboard's network access is scoped only to those moments. The Terms and Privacy Policy are the legal boundary, but independent security audits or code-level verification were not found in the available sources.

Broader AI Industry Trend: Embedding Agents into Mobile Surfaces

Acti's launch is a concrete example of a major 2025–2026 industry shift: embedding AI agents directly into existing user interfaces rather than requiring users to open a separate chatbot app.

• Gartner data – 80% of enterprise applications shipped or updated in Q1 2026 now embed at least one AI agent, up from 33% in 2024 .
• IDC forecast – AI copilots are expected to be embedded in nearly 80% of enterprise workplace applications by 2026 .
• The "agentic leap" – Google Cloud and other analysts describe 2026 as the year AI shifts from answering questions to orchestrating end-to-end workflows semi-autonomously .
• Keyboard as the insertion point – Acti takes this trend to the mobile keyboard specifically, arguing that "ChatGPT and Siri are places you go; Acti is already where you're typing" . This mirrors the broader strategy of turning the lowest-friction surface (the text field) into the agent's entry point, rather than creating a separate app or overlay.

In short: Acti is a Singapore-launched, AI-native keyboard that replaces predictive typing with intent-to-action execution, addresses privacy with a "send only on explicit trigger" design, and fits squarely into the 2026 industry push to embed autonomous agents into the interfaces users already use every day.