Acti

把 AI 智能体做进手机键盘,在任意输入框里长按即可取信息、跑动作、触发第三方工具,不必切 App

In-depth Report

  • Acti (meaning "action") is an "agent keyboard" from Singapore. It will be launched on both iOS and Android on June 30, 2026. Its core idea is very straightforward: build the AI ​​​​intelligent directly into the keyboard you use most every day, so that you can "get information, run actions, and trigger tools" in the input boxes of chat, email, and social applications, without having to switch out and open another AI App. The bottom layer is driven by the Google Gemini model. It topped Product Hunt on the day it was released and completed a $5.3 million seed round led by BITKRAFT Ventures. Its imagination lies in "redefining the basic interface of the input method", but the risks are also concentrated - a keyboard with networking and third-party authorization capabilities is naturally in the bull's-eye of privacy and security disputes.

  • Acti is developed by OpenActi, a company headquartered in Singapore. Founder and CEO Young Wang has been at Baidu for nearly ten years. He once built Facemoji Keyboard to have more than 300 million daily users and has a deep accumulation of input methods. His public expression is: when LLM appears, text is no longer just "something to be entered" but becomes a "carrier of intention", and the keyboard should be redone. There are also two key figures in the team: CTO Mike Sun is the founding technical director of Baidu Cloud Photo Album Yike Album (the business has more than 10 million daily users), and CSO Junbo Yang comes from the encryption investment institution HashKey Capital. Such a team shows that this is not a weekend project for two people, but a team of veterans with first-hand experience in input methods, cloud services and consumer investment. At the financing level, Acti announced the closing of a $5.3 million seed round at the time of its release. Jonathan Huang, a partner at leading investor BITKRAFT Ventures, said that they were betting on "the next stage of human-computer interaction." For a keyboard application that has just debuted, this start-up resource is quite substantial.

  • The protagonist of Acti is "ActiBar", which replaces the original space bar: pressing it normally means typing, and pressing it long will trigger an action. For example, if a friend asks you where you are, you type the location in the dialog box and long press ActiBar, and it will insert the location directly into the chat box; when it comes to a certain stock, it can post the real-time price; when it comes to eating, it can drop nearby restaurants or reservation links into the conversation. It will even generate interactive "mini-apps" that allow the other party to browse what it finds directly in the conversation. More than just the space bar. You can map any key to a "Skill Key" for multi-step tasks: long press T to translate a message, long press C to generate a meeting link, long press N to bring up a Notion document, long press L to pull up someone's LinkedIn profile, long press M to generate a Google Meet link. These skills do not require writing code - you describe what you want in plain English, and Acti will use Skill Builder to help you build it, which can be used privately or published to the community Skill Hub for sharing. Before the official release, beta users had created more than 1,000 skills in less than two weeks, covering real-time World Cup data, Polymarket prediction market links and other scenarios. The product emphasizes a local-first privacy model: personal context remains on the device by default, and text will only be sent for processing when you actively trigger a function that requires an Internet connection. When connecting to third-party services, OAuth authorization can be done in advance in one go, or you can apply on-demand when skills are triggered, and only access the tools you authorize.

  • Acti is currently available for free. It's marked as free on the App Store and Product Hunt, and Google Play shows it has been downloaded more than 100,000 times with an "in-app purchase" mark. AI capabilities manage usage through an internal Credits system. Official terms currently state "No paid subscriptions, no in-app purchases, no paid plans, and no paid Credits," but TechCrunch reports that the company plans to launch subscriptions in the future with stronger models, higher daily limits, and more advanced features. In other words, free is now priced during the customer acquisition period, and the commercialization path has actually been drawn up, but it has not yet been implemented. A noteworthy product action: In the Product Hunt comment area, team member "William Wang" repeatedly used the password "Acti Code to unlock LIFETIME Premium" to reply to supporters. This is not only a new gimmick, but also implies that payment forms such as "lifetime premium version" are already in preparation.

  • Early feedback is clearly divided into two camps. On Product Hunt, picky PH users gave a rare warm response, with Acti taking the top spot of the day, with a score of about 423–535. User nicole_h94's comment has been quoted many times by the official: "Unexpectedly, I use Acti more for "retrieval" than "generation" - I rarely let it write long articles, mostly for cache or quick execution." This evaluation point highlights the true value of the product: it is not another writing assistant, but compresses "getting things and doing things" into the action layer of the input box. User call_me_young (founder Young) also said that many people underestimate the friction of cutting apps and don’t realize it until they use Acti. Another high-frequency praise is that "long press to get calendar/meeting links is smooth" and "you can pull up real-time scores and nearby restaurants without leaving the conversation, it's like magic." However, some people pointed out the threshold for getting started: the entrance to the custom workflow is not conspicuous enough when it is first launched; when multiple applications can respond to the same vague command, Acti defaults to the calendar and still requires you to set it first, and the experience is not smooth enough. The rating sample is too thin: the App Store shows 3.7 stars, but only 7 ratings, which is far from enough to judge; there is no overall rating on Google Play yet. AndroidGyaan mentioned that the iOS side currently only supports the English interface, and the keyboard itself supports typing in 34 languages, but the company did not provide a complete list.

  • The technology media generally considers this idea “promising.” TechCrunch reported the release and financing in detail, and cited the founder’s judgment on “context fragmentation”—the fundamental limitation of today’s AI agents is that user context is fragmented in different apps, and Acti spans all these apps, so it can build a context layer that “truly belongs to the user, not to the platform.” Technologist Mag wrote after the actual test: "After watching it run, I think the idea is tenable." Digital Trends framed it as "living the AI ​​assistant into the keyboard, so you never have to open the app." Tool navigation sites such as aipure define Acti as "the world's first proxy keyboard" and break down its core selling points: cross-application command layer, customizable skill keys, code-free builder, and local-first privacy. But there are also calm voices. The Indian Express poses an open question: Does the average consumer really want to “put a smartphone directly into the keyboard”, or do they prefer a standalone app or system-level assistant? No one can fully answer this question yet.

  • Privacy is an unavoidable minefield. Any third-party keyboard that wants to truly connect to the Internet, run AI, and connect to third parties on iOS must turn on the "Full Access" permission. The Acti team confessed in the comment area: Without Full Access, the keyboard can basically only be used for typing and cannot be connected to the Internet, so this permission is indeed required except for basic input; they emphasize "data minimization" and "only access when you explicitly trigger it." But blog.heyeveryone.io points out the contradiction sharply: on one side, the company says "we do not record keystrokes and do not upload content", and on the other side, "this keyboard requires full network access in order to work" - the tension between "the company says no keystrokes are recorded" and "the keyboard inherently requires Internet access" is what makes people uneasy. A bigger hidden danger is irreversible operation. Acti's terms clearly warn: AI output may be wrong, incomplete, biased, or out of date; when a Skill calls a connected tool, AI may misread the intent, select the wrong recipient, or insert the wrong content. The terms specifically remind users to review the actions recommended by AI "before sending or saving", especially irreversible operations such as sending messages, deleting records, and sharing files. Put an AI agent that "thinks first and think later" in your most casual and fastest-speaking chat input box. This is a risk that most people won't think about before something goes wrong. There are also two structural risks. First, keyboards are the “private territory” of Apple and Google. If proxy keyboards really become a trend, Gboard, Apple’s native keyboards, and Samsung keyboards are entirely likely to absorb this idea, and third-party players will only be able to grab the leftovers. Secondly, capabilities are highly dependent on the richness of integration - the number of skills and the breadth of connected services directly determine whether it is useful; once the ecosystem fails to take off, the ingenuity will come to nothing. Another fact to point out separately: When I visited acti.app during research, the domain name landing page displayed the GoDaddy "sale/lease" page. This is inconsistent with the narrative that media such as TechCrunch regard acti.app as an official domain name. It may be due to domain name transition, configuration error, or brand migration. However, as a user to judge "where to go and where to look," this signal is worth alerting. It is recommended to refer to the official account on the App Store / Google Play / Product Hunt.

  • Suitable people: Professionals who spend all day long in messages, emails, and various text boxes; roles in sales, operations, marketing, etc. that need to "quickly retrieve information, generate replies, and initiate schedules in conversations"; and people who no longer want to interrupt their flow with the action of "opening an app—copying—switching back—pasting." Its value is not in "helping you write", but in "helping you take and handle it". Not suitable for people: People who are extremely sensitive to privacy and are unwilling to give full access and OAuth permissions to third-party keyboards; people who only work on the desktop and rarely use mobile phones to type; and enterprise-level automation scenarios that require auditability and traceability - currently it is more like a personal efficiency toy, and the boundaries between responsibility and audit are still blurred. Substitutions and supplements: If you value "not leaving the conversation" but accept independent apps, Perplexity, Gemini's floating windows or system-level assistants are lighter options; if you want to do complex cross-application automation, mobile phone shortcuts or desktop agents are better in controllability. The unique point of Acti is "welding the action layer to the input source", which is a wedge that others do not have.

  • Acti has reimagined an interface that has been neglected for twenty years - the keyboard - as the "action layer" in the AI ​​era. The direction is right, and the initial potential and team background are also excellent; but its real test question is not the functional demonstration, but whether the user is willing to leave the Internet and third-party authorization to us, and irreversible operations to AI. Whether it can establish long-term trust. The domain name implementation is abnormal, the privacy model is still at the "commitment" level, and the scoring samples are too thin. These are all question marks that must be raised at the moment.

User Reviews

  • 头像
    BeverlyFoster_99
    用了一周Acti了,最香的场景就是查餐厅推荐这件事。朋友在微信问附近有啥好吃的,长按空格键它直接弹出评分和地址,不用切大众点评了。

  • 头像
    Raymond.Scott_Max
    说实话我对AI键盘本来是有PTSD的,之前用过好几款都不太行,要么隐私协议写得像恐怖小说,要么生成的内容土得掉渣。但Acti还真不太一样,它不是只给你推荐下一句话,而是真的在帮你「做事」。看完TechCrunch那篇报道我决定再给AI键盘一次机会,这次没失望。

  • 头像
    ELeeQ
    Acti绑定的T键翻译功能太好用了,跟外国同事聊天长按一下直接整句翻译完塞进对话框,省了来回切DeepL的功夫。

  • 头像
    海浪549
    Acti 的想法确实新颖,但我还是不太放心把键盘权限交给第三方公司。iOS 上要开 Full Access 才能用,那个警告弹出来的一瞬间我就犹豫了。

  • 头像
    Natalie.Simmons_X
    太香了,回不去了,谁用谁知道。

  • 头像
    ElaineHenry
    Acti 底层用的 Gemini,响应速度确实快。不过我试了几个中文场景的 Skill,偶尔会有理解偏差,毕竟是英文优先训练的模型。如果你主要用英文的话体验会顺畅很多。

  • 头像
    MelissaCollinsJr407
    Acti 在 Product Hunt 上拿了 500 多票,看了下评价两边倒。说好的觉得它是交互革命,说不好的担心键盘变数据漏斗。我属于观望党,先装了一直没深度用,等它生态再成熟一些再说。

  • 头像
    流年_12
    不支持中文语音指令是硬伤,期待更新。

  • 头像
    95pa9t88
    早上设了一个 Skill,把 M 键绑了 Google Meet 链接生成,现在开会前再也不用手动去日历里复制链接了,直接长按就出来。这种小效率提升一天能省好几次切换,积少成多还挺明显的,推荐大家都试试。

  • 头像
    Kyle836
    看到有人说 Acti「不适合长文写作和合同判断」,我深表赞同。它的最佳场景就是轻量快速的操作,别指望用它写论文。日本用户nikki的评价很克制——它适合移动中的搜索、定型备忘和短确认,但不适合长篇内容创作和需要精确判断的场合。

  • 头像
    trueMikaelKetola_88
    少切一次就是一次,真的香。

  • 头像
    DonaldKruger
    刚刚试了用 Acti 查实时股价,在群里聊股票的时候长按一下,价格直接显示在聊天里,朋友们都问我装了什么插件。这种感觉就是——你还在切 App,我已经发完了。

  • 头像
    JohnPedersen
    有一点必须吐槽:iOS 上第一次使用各种 Skill 时权限确认框弹个不停,翻译要授权、定位要授权、读取剪贴板又要授权,一连串点下来体验确实不好。能不能把权限申请整合成一次性的设置向导?

  • 头像
    Lauren.Hall_66
    Acti 创始团队背景挺硬的,百度 Facemoji 键盘出来的团队,3 亿日活的产品经验不是白给的。难怪能把交互做得这么顺手。

  • 头像
    AnnStephens520
    我在对话里用得最多的是取日历链接,比切 App 顺多了。

  • 头像
    Amy486
    我比较在意隐私问题。官方说本地优先,默认不上传数据,但它是第三方键盘这件事本身就让人有点不安。我目前只用来查不敏感的信息,涉及密码和金融信息的时候还是切回原生键盘。

  • 头像
    ArthurFisher50
    终于有个键盘不是只会纠错,而是能办事了。

  • 头像
    Judith.Myers_2023
    域名 acti.app 现在居然是出售页,有点迷。媒体都把它当官网写,结果落地是 GoDaddy 租售,建议下 App 还是认准商店和 PH 官方号,别被山寨页带偏。

  • 头像
    DAllen_2024380
    最大的隐忧是苹果和谷歌。Gboard 要是哪天把这套「长按出动作」抄进原生键盘,Acti 这种第三方基本没活路,毕竟没人会为了一个输入法换掉系统默认。

  • 头像
    JTorres_77
    Acti 把空格键变成 Agent 这个设计太聪明了。正常按是空格,长按是行动。不用学新的交互方式,肌肉记忆就能覆盖。

  • 头像
    剑客_4
    它说本地优先、不上传内容,但底层是 Gemini,真触发时数据总得出去,这个边界讲得有点含糊,希望后续有更透明的说明。

  • 头像
    Jason_Allen_Pro
    隐私这块我还是观望。键盘要联网、要 OAuth,等于把最随意的聊天入口交给 AI,万一它误读意图发了不该发的内容,撤回都来不及。条款虽然写让你复核,但真用起来,大多数人不会每次都看。

  • 头像
    ChristianAdams_66
    创始人做过 3 亿日活的输入法,底子不虚。

  • 头像
    Ralph.ReyesIII
    长按 ActiBar 把定位丢进聊天框那一刻,真的有点魔法感。

  • 头像
    尹萱霖
    真正让我留下来的是「检索优先」而不是「生成优先」。我一天到晚在微信和邮件里,以前查个会议链接要跳出三次,现在长按一下就完了,这种微小的摩擦消解累积起来很可观。

  • 头像
    Denise.GutierrezQ
    Acti 免费的,这点让我愿意试一下。

  • 头像
    Jack_ThomasK87
    长按 M 生成 Meet 链接很顺,远程会议前不用再翻日历 App 了。

  • 头像
    Nathan_Rivera
    整体方向我看好,键盘作为「动作层」比再装十个 AI App 都合理。就等它把权限和尊重用户复核这两件事讲清楚、做扎实。

  • 头像
    Sara.RodriguezJr
    技能键映射自由度很高,我把常用那几个键都改成了自己的工作流。

  • 头像
    Danielle.Roberts0071
    内测两周社区就造了 1000 多个技能,这个生态势头挺猛。不过技能质量参差不齐,有些明显是凑数的,Skill Hub 后面得有筛选机制,不然新手一进来就被垃圾技能劝退。