App Review | Recent Apps Worth Watching: Local Code, Hoardware 3.x, Vimpo, Klick'r, Blo…

English text is being prepared.

Click the button below to copy the link

Preface

Welcome to this issue of App Review. You can use the table of contents to quickly jump to the content that interests you. If you've found other interesting apps or topics you'd like to follow, feel free to discuss them with us in the comments.

English text is being prepared.

How can you run large models on your own computer? Those with good hands-on skills may be used to downloading and managing models through Ollma and then calling them locally, but for ordinary users, is there a simple and easy-to-understand local model deployment solution?

Local Code is a local-first AI Agent tool for the Mac platform. Compared with the professional model deployment tool Ollma, Local Code focuses on providing the optimal solution for Mac devices—the app offers only three local models in total, and during initialization it matches the most suitable local model based on the device configuration. For example, my MacBook Air with an M4 processor can only choose the lowest-tier Qwen 3.8 27B because it has only 16GB of RAM. After choosing a model, downloading the model files, and completing setup, you can enter the main interface of Local Code.

In terms of functionality, Local Code is not much different from mainstream AI Agent tools on the market. You can start conversations through project folders in the sidebar; conversations support adding images, documents, and other attachments, and you can also interact by voice. On the far left, you can also fine-tune plugins—for example, in settings you can choose whether to use an on-device model or a cloud model, and the reasoning intensity can be adjusted according to the machine's own performance. Here, because my performance is weak, I directly set it to low; as for the thinking budget, it likewise needs to be selected according to the local machine's performance. The plugin settings also include options for whether to enable certain permissions or features. To be honest, compared with mature commercial AI Agent tools, the gap may be more than a little.

In the chip button in the sidebar, you can also adjust the model, such as whether to prewarm, how long after idle to unload the model, and whether to use the Apple Neural Engine for 8-bit S models, etc. If you're not sure, just keep the defaults.

In terms of actual model throughput, because my device configuration is average, the experience is barely satisfactory. Using high thinking mode, a simple GitHub project analysis may take nearly 1 hour, and the token generation speed is almost in the single digits; in simple text summarization tasks the speed is slightly faster, but the token generation speed is still only in the tens. It can be seen that running large models based on the NPU capabilities of consumer-grade hardware simply cannot reach the output speed of today's commercial cloud models.

Of course, if your Mac has higher performance and can even run the quantized DeepSeek 4.1 Flash 1017, then my above experience is not a useful reference.

Local Code gives those who want to try running large models locally but don't know how to deploy, download, and handle other complicated processes an easier-to-use platform, and its operation is fairly foolproof. However, for low-end Macs, experience problems caused by the overall computing power bottleneck still exist, and those with average configurations are advised to try it mainly for novelty. Local Code adopts a business model of free local operation and separate charges for cloud models. If you're interested, you can download it from its official website.

Hoardware 3.x Update: Adding Statistics Displays for Collections, Turning iPhone Duo into a Display Case

Platform: iOS, macOS, iPadOS

@Vanilla: Not long ago, I introduced Hoardware in App Review, an Apple hardware collection archive based on AppleDB created by independent developer Aaron Perris. Less than three weeks after the last introduction, Hoardware quickly updated to version 3.1. This time we'll take a look at the new features of both versions 3.0 and 3.1 together.

The star of version 3.0 is Hoardalytics, which, in addition to the existing system coverage, organizes the entire collection library into a set of interactive charts where you can view color distribution, device release timelines, and model coverage; the statistics can also be turned into collection cards, making it easy to share an overview of your collection with fellow enthusiasts or post it to social media.

On the physical management side, version 3.1 adds asset tags, with customizable prefixes, numbers, and categories, and lets you preview and print labels bearing barcodes, serial numbers, and an optional logo; once affixed to physical items, they make it easy to match devices with the numbers in your records. Files such as invoices and purchase receipts can also be stored alongside devices, synced via iCloud, and included in library backups.

In addition, version 3.0 adds an upgrade history that can record the path of device upgrades and after-sales replacements.

When your collection is large, you can also just ask Hoardini—this AI assistant, added in version 2.0, can answer questions about your collection, device specs, firmware history, and software compatibility, and supports follow-up questions; its answers also include links to the corresponding devices, which you can click to jump directly to the relevant device detail page, and conversation history can also be synced via iCloud; version 3.0 also integrates with Siri and Shortcuts, so you can find devices, read device details, check for updates, get a collection summary, or ask Hoardini directly.

Note that asking questions through Siri and the Ask Hoardini action in Shortcuts requires a device that supports and has Apple Intelligence enabled.

For the iPhone Duo, Hoardware completed basic adaptation first in version 3.0: when expanding or resizing the window, the currently selected device and any content being edited are preserved. Version 3.1 then adds the more playful Hoardshelf: when you set the Duo in tent mode, it displays a clock and rotates through devices in your collection, and you can also enable physical effects that change as the phone tilts. This feature also respects the system's Reduce Motion setting.

Version 3.0 also improves the model selection experience: groups now show the number of devices, and the wishlist is now editable; on iPhone and iPad you can also update the firmware version recorded in your archive, or mark an entire group of devices as up to date. The scanning feature added in version 2.0 can read another device's "About" page, reducing manual entry when creating records.

Overall, for users already using Hoardware to manage their collection archives, these updates make it more convenient to view the composition of a collection, manage physical items, and keep receipts, and users with an iPhone Duo also gain another way to display their collections. Hoardware is currently available for free on the App Store, with no ads, in-app purchases, or subscription options; it requires iOS or iPadOS 17.0 or later and macOS 14.0 or later, and the old devices being cataloged do not themselves need to be able to install the app.

Vimpo: Video Downloading, Bilingual Subtitles, and AI Editing All in One

Platform: macOS

@ElijahLee: Vimpo is a video processing tool for macOS that integrates video downloading, speech recognition, AI translation, bilingual subtitles, and video editing. It supports downloading videos from over a thousand sites including YouTube and Bilibili, and can also import local videos, use an on-device speech recognition model to generate subtitles, and then use large language models to complete translation and content analysis.

Vimpo's video download feature is built on the open-source project yt-dlp and supports over a thousand sites including YouTube, Bilibili, Vimeo, Twitch, and TikTok; in Vimpo, you just paste a video link to download the video and save it locally. After downloading, videos go into a unified library, and in addition to online videos we can also import local files, with the option to copy them into the library or link directly to the original file.

One of Vimpo's core features is automatically generating bilingual subtitles. Traditional video subtitle tools usually need to rely on the video's existing subtitles or upload the video to the cloud for speech recognition, whereas Vimpo combines on-device speech recognition with translation by external large language models. It has a built-in Whisper speech recognition model that can run directly on the Mac and convert speech in videos into text; because the video does not need to be uploaded, the local approach also supports offline recognition.

The accuracy of Whisper speech recognition is limited by model size and accuracy, and Vimpo offers model options in various sizes, recommending a suitable model when we download a video. We can also go to the app's Settings - Speech Recognition to view the size, speed, and accuracy of the built-in models and choose between recognition speed and accuracy according to our Mac's hardware performance. By default, the recommended Large v3 Turbo (compressed version) can handle recognition tasks fairly well, though errors occasionally occur, such as recognizing iPhone Duo as iPhone 2.

After generating the original subtitles, Vimpo can call large language models to translate the subtitles into the target language. It supports services such as OpenAI, Anthropic, Gemini, DeepSeek, Qwen, Moonshot, Zhipu, and MiniMax, and is also compatible with local models such as Ollama and LM Studio. We only need to configure our own API Key to complete subtitle translation.

Vimpo's intelligent subtitle processing features, including smart sentence segmentation, name correction, and terminology consistency, can make subtitles for long videos easier to read. During playback, the original text and the translation can be displayed simultaneously, and styles such as font, color, and position can be customized.

Vimpo calls the video playback interface the "Theater," where we can watch the video, read bilingual subtitles, and view the full transcript at the same time. Clicking a sentence in the transcript causes the player to automatically jump to the corresponding timestamp. This design links video watching and text reading together. When encountering content we do not understand, we can locate it directly through the transcript without repeatedly dragging the progress bar to search.

Another interesting feature is AI highlight detection. For example, for a one- or two-hour long video, AI can analyze the important viewpoints in the video based on the subtitle content, automatically select segments worth paying attention to, and generate a title, time range, and recommendation reason for each segment.

In addition to watching videos, Vimpo also provides a "Studio" for content creators; professional users should find this window familiar. Here, features such as the picture, subtitles, and tracks are all laid out according to the layout of video editing applications. In "Studio," by pressing I/O to set in and out points, we can create a new clip and manually extract the highlights from a long video to make independent short video clips.

As of publication, Vimpo offers a macOS Beta test version, which you can download from the official website. During the test period, all features are free. The official version is planned to use a one-time purchase model rather than a subscription.

English text is being prepared.

The Klick'r introduced today adds recognition of screen content, giving clicker-type tools the ability to "see first, then execute."

Compared with traditional clicker tools, Klick'r's biggest difference is that it does not use "time" as its execution logic. Instead, it uses "events" similar to IFTTT or Tasker as judgment conditions. In smart mode, you first need to enable the floating window and select a button, icon, or other identifiable area on the screen as the reference target; if necessary, you also need to set a tolerance value.

During execution, only when Klick'r captures content or text on the screen similar to the target you just selected will it execute the click, long press, swipe, and other operations you created. For example, we now often check in in various apps. If the network is poor, a traditional clicker will just randomly tap on a check-in page that has not loaded, while Klick'r will wait until the "check-in" graphic appears on the screen before clicking.

In addition to general clicking features, Klick'r also has built-in workflows that support logical judgment, supporting both AND and OR condition combinations, and provides advanced actions and triggers such as counting, app jumping, event states, and system operations. We can combine a series of conditions; for example, when completing tasks in a mobile game, if Klick'r detects a specific "collect reward" button, it executes the action "click collect reward" 10 times.

A tool like Klick'r that supports visual recognition and advanced condition combinations reminds me of Tasker, which I have not used for a long time. Although both have some learning curve, once you are proficient, the degree of automation will jump significantly. If you want to learn some complex features, you can directly view the help documentation in the app and learn how to use them through the guided process.

Of course, Klick'r also supports the traditional mode of fixed operations by time and coordinates. If you do not need advanced features for now, using Klick'r as a free screen clicker is also a good choice. You can download Klick'r for free on Google Play, F-Droid, or GitHub.

@Peggy_: After years of fierce competition, note-taking apps have basically formed a relatively stable competitive landscape. Users who like to research and store various materials can use knowledge-base-type note-taking apps that support bidirectional links, while users who like to produce content and are not strict about formatting mostly use various note-taking apps that support plain text format well. Further subdivided, whether network synchronization is supported and whether cross-platform use is possible... are all important reference factors when users choose a note-taking app.

The Bloom app we're introducing today has plain text format as its primary feature, allows fully local operation, and can switch platforms seamlessly. If you're still interested after reading this, you might want to dig deeper: when it comes to plain text, most people still think of the txt format or the md format that has become popular in recent years. However, beyond these two, there is another plain text format: the org format. Like the md format, it develops a special markup language on top of the original plain text format, so the text looks formatted when previewed. The advantage of both is obvious: you can write without being distracted by formatting.

The difference is that the org format was created from the start to solve problems of task management and information organization, and its syntax support for tasks and to-do items is more complete. Bloom also happens to be a note-taking app with native support for the org format. After opening the app, Bloom provides three built-in templates: diary, quick note, and task note. We can modify or add templates according to our note-taking needs.

Take task notes as an example. Users familiar with the relevant syntax can directly create a new file, and common syntax options are arranged above the keyboard, further eliminating the need to look up symbols yourself. If you are not familiar with the relevant syntax, you can still use Bloom as an ordinary note-taking app. When you need to add or manage personal tasks, tap the menu button in the lower right corner of the editing interface to add various attributes to a task, such as task status, priority, tags, and even time attributes. Later, as you use it more often, the syntax will no longer be a problem.

As the amount of written content increases, tasks scattered across different notes are automatically organized by the app. Tap the sidebar button on the main interface to see all tasks or the details of tasks within this week. The quick search content here can also be customized—after all, all files in Bloom are written in plain text, so search naturally supports plain text plus relevant syntax. If you often search for a certain type of specific content, you can also save the corresponding search keywords for direct reuse later.

In terms of security, you can choose to save all content on your phone, or use cloud sync software yourself to back up the corresponding local folder to the cloud. Cross-platform use is equally simple and easy: all files you write in Bloom can be opened and further edited on other platforms through editors that support the org format.

You can download and try Bloom via F-Droid. The app is open source and completely free.

Chunchun Writing: Well-known writing app officially launches iOS version

Platforms: iOS/Android/Linux/Windows/macOS

Keywords: writing, all platforms, text editing

@Peggy_: A week ago, the well-known Android writing app Chunchun Writing was officially listed on the China region App Store.

Chunchun Writing has always focused on a writing experience that "never loses" your work. It had already been continuously updated on Android for several years and gradually expanded a desktop version to improve the editing experience. The launch of the iOS version finally completes this large piece of the puzzle for mainstream platform support.

The biggest difference from the Android version is that this iOS version of Chunchun Writing fully embraces liquid glass design. Common function buttons on the main interface are presented in liquid glass, showing a very high degree of platform adaptation.

In addition, since it has come to iOS, support for syncing via iCloud is a natural step. We can look forward to Android also supporting iCloud sync later to further improve the sync experience.

On top of the features already available in the Android version, this update is also the first to support a two-column mode, meaning you can open two editors at the same time and edit side by side. This is great news for iPad or iPhone Duo users, though it still feels a bit cramped on a regular iPhone. This feature will also come to Android later.

The most practical customizable quick input bar is equally useful on iOS. In the past, in various editors, if we wanted to enter double quotation marks, we either had to type them manually in two steps, or first enter the symbol and then move the cursor to add text. Chunchun Writing supports custom cursor positions and can, after input is complete, move the cursor directly to the end of the symbol with the Enter key, greatly improving text editing speed.

Overall, all of Chunchun Writing's current editing experience has been perfectly replicated on iOS, but the time machine feature and AI features are not yet available. If you have previously purchased a Chunchun Writing membership, you can also log in directly on iOS and sync your membership status without paying twice. In this respect, Chunchun Writing can be called a conscience of the industry.

If you are a loyal Chunchun Writing user who wants to continue the "never lose" writing experience, or if you simply want to try a new app, you might as well download it from the App Store and give it a try. The app's basic features are free.

Photon Cam: Unlocking new camera features for older iPhone models

Platform: iOS

Keywords: camera, film, manual control, filters

@Snow: In iOS 27, Apple brings new texture controls to the photography features of the iPhone 18 Pro and iPhone Duo. Open Photographic Styles, and while changing the color palette, you can adjust soft skin, glow, and film grain effects in the Texture section—I thought this update would leave a batch of third-party apps "unemployed," but unexpectedly Apple "struck another blow": this feature has not rolled out to all models with the system update. Only photos taken with designated models such as the iPhone 18 Pro series can use these new control options.

So you can't blame third-party cameras for stepping up. For example, PhotonCam, in its latest version, lets older models enjoy soft skin, grain, film, and other effects too.

After version 1.5 "opened up" the color palette for all models, version 2.1 of PhotonCam successfully unlocked texture controls. Tap the top-right corner of the main screen to enter "Camera Settings - Lab" and turn on the switch; photos taken and saved with PhotonCam will no longer be limited by device model. As long as you meet the iOS 27 system version requirement, you can use the latest texture controls in the system Photos app. Just as with the native camera, you don't have to rush to decide the look of the image while shooting with PhotonCam—texture effects such as soft skin and grain all support later adjustment. PhotonCam itself only participates in the feature switch and the shooting process, and will not significantly affect your existing editing workflow. Moreover, you can use PhotonCam to import photo material not shot on an iPhone and force-enable Photographic Styles, further broadening the scope of this feature.

In the 2.0 update two weeks ago, PhotonCam also brought a starburst effect to models without a variable aperture within the app. You can adjust parameters such as the point light source threshold, light effect intensity, extension length, number of starburst rays, and starburst color temperature in "Photo Editing - Adjust - Effects - Starburst." Although it is certainly not as natural as hardware-backed effects, nor can it help you avoid flare and ghosting, it can still add some atmosphere to your night photos.

Setting aside the "early taste" of new features, PhotonCam itself is also a fairly complete manual camera. It offers shooting options such as shutter speed, ISO, color temperature, focus, RAW, and ProRAW, and also supports post-processing tools such as LUTs, curves, distortion correction, and borders. Version 1.48 supports adjusting the hardware-side variable aperture of the iPhone 18 Pro series in manual mode. Even if you hold the latest device, PhotonCam can help you optimize the manual shooting and post-processing experience.

PhotonCam includes an in-app purchase service: you can subscribe for 28 yuan/month or 78 yuan/year, or pay a one-time 98 yuan to purchase. After paying, you can unlock features such as HEIF/JPEG photo watermarks, all manual control items, advanced editing tools, batch LUT import, and texture controls. If you need a sufficiently professional manual camera and want to try the new texture features on an older model, PhotonCam will be a good choice. Note that PhotonCam does not currently support enabling Photographic Styles for Live Photos. If you frequently shoot Live Photos, or are looking for a video shooting tool, it cannot meet your needs. 1

> Download the SSPAI client and follow the SSPAI official account to unlock a new reading experience 📰

> Practical, useful genuine software, presented to you by SSPAI 🚀 App Recommendation Review Review This article's editor:

@Clyde

5 SSPAI friends have already charged