One of the biggest updates that landed with macOS Tahoe was a turbocharged Spotlight. Apple fixed a bunch of papercuts and added new perks that will appeal to power users. Of course, the usual “sherlocking” happened. But not all the changes were well-received.
For instance, the death of the classic LaunchPad drew a fair bit of criticism, and spawned a whole bunch of apps that bring back the experience. Likewise, the Spotlight upgrades also borrowed heavily from productivity tools such as Raycast, and they were not as universally praised as third-party alternatives.
Personally, I find the new Spotlight experience a little overwhelming, and somewhat functionally hollow at the same time. This is where Vector comes into the picture. It’s a minimalist Spotlight replacement developed by Ethan Lipnick — a former engineer at Apple — and offers some really cool AI-driven conveniences.
What does Vector do?
At the most fundamental level, Vector wants to replace the broad role of Spotlight. Just like Spotlight, it lives as a dedicated app icon in the dock, but you can also access it from the menu bar to reduce the dock clutter. So, what can it do?
Launching an app, for starters. But instead of opening the full Spotlight window that hogs the central space on the screen, you can launch Vector from any corner of the screen to keep things tidy and visually non-intrusive.
Talking about flexibility, you can set custom keyboard shortcuts to pull up the main panel, clipboard mode, or emoji picker. Likewise, you can pick the most convenient keyboard combo for starting a file search, or look through your chats in the built-in Messages app.
To Vector’s credit, it not only looks extremely cool, but the clean design and snappy animation make it feel like something designed by Apple. It actually feels faster than interacting with Spotlight, and its search performance also feels nearly as fast as what you get with Spotlight.
I noticed that the semantic search system (especially for the files stored on your system) in produces faster results. The only caveat is that when you are looking up a file in the Vector search window, you don’t see a preview.
So, if you’ve saved a series of files with names such as ABC-1 and ABC-2, you are essentially in a blind spot. Another minor hiccup is the search performance. By default, Vector runs a local AI model that is only 64MB in size. Its search performance, however, is not nearly as good as what Spotlight can do.
For example, when I entered the flight number in the search field, Spotlight automatically surfaced the boarding pass on which the number was printed, but Vector failed. If you want better semantic search output, you will have to download the more powerful BGE-M3 model, which takes 1.1GB of space.
Lots of hits, a few misses
The indexing process is rather opaque, even though it appears to work fine with the files stored on the system. For example, I could not reference my most recent chats with friends and family members on the same date. But random service messages and code returned a valid result when searched within the Message directory.
The semantic understanding is also a hit or miss. For example, when I look up “Definition of catharsis,” I get results pulled from the Dictionary app and Wikipedia. However, when I try a contextual search for content within PDF files, Vector would fail.
It worked well with pulling up forecast information from the weather app, but faltered at pulling details from the Calendar app, even when the queries were straightforward. Finding entries that should lead to a Maps view was a fairly reliable experience, but as soon as you dig into natural language queries such as “distance between Umpling and Laitumkhrah,” it stuttered.
Vector dips into a wide range of sources to answer your queries. The list includes everything from Calendar and Dictionary to Contacts, Maps, Weather, Wikipedia, apps, messages, files, and even the emoji deck. You can disable the indexing (and semantic search system) on a per-app basis in Vector. I like this flexibility, as it not only ensures control over privacy but also reduces the processing load.
On the positive side, I absolutely love the clipboard system. When you summon it, you get a sliding carousel of cards that feels buttery smooth to slide through. Another neat touch is that each card also shows the app from which the content was copied, alongside the date and/or approximate time.
Vector offers plenty of flexibility in terms of how you interact with the app. You can use it solely as a full-fledged app launcher, deploy it as a system-wide semantic search system, or merely dig into its built-in clipboard chops. Additionally, you can select between six positions where the Vector window opens.
I kept it anchored to the lower right corner, as it looks pretty neat and doesn’t disrupt the view of foreground app windows. The clipboard system also lets you set an auto-delete protocol ranging from a day to a full week or month. You can choose to keep everything in the clipboard directory forever, as well, without worrying about security, as all the copied-and-pasted content is processed and saved only on the device.
Reading the room (for the silicon inside)
I have repeatedly written that Apple’s silicon is in a league of its own. Whether it’s Macs or iPhones, the cumulative balance of raw performance and efficiency is miles ahead. But despite that lead, Apple doesn’t let you tinker with the performance output.
On a Windows machine, you get native utilities like Armory Create (on Asus ROG machines) and third-party apps that offer granular control over everything from GPU frequency to fan speed. Even phones such as the Red Magic 10S Pro and OnePlus 15 let you tap into the real potential by adjusting the performance presets.
Vector doesn’t quite fix that situation for your Mac, but as long as you’re running the app, you can adjust its performance output. To start with, you can choose to run the AI-powered functionalities in vector solely atop the CPU. But if you want better performance, you can combine the CPU and the GPU, or push the CPU and the neural chip (NPU) together for faster output.
And if you’re not worried about the power draw, the app also lets you separate the workload across the CPU, GPU, and the NPU simultaneously. Apple’s M-series processors come with a fairly powerful neural engine, so the best combination for running vector is separating the workload across the CPU and NPU.
In case you have a beefier flavor of the silicon, such as the M4 Pro or M4 Max — both of which pack more GPU cores — it’s worth selecting the performance profile that also puts the GPU into the mix. Based on the chip fitted inside your Mac (and the kind of performance you are chasing), you can make a few other modifications.
To begin, you can pick between the BGE-Small model, which is just 64MB in size and runs by default. This one is plenty powerful for contextual search across the local files container and messages. However, if you want better responses and support for more languages, the BGE-M3 model is where you should look at.
M3, here, stands for multi-functionality, multi-linguality, and multi-granularity, which is pretty self-explanatory as far as the model’s benefits go. This one takes slightly over 1GB of space, but performs much better at retrieval of contextual information, and supports longer inputs worth around six thousand words (8192 tokens).
You can separately set the speed at which content in the Messages app and Files explorer is indexed. The app runs fully offline, which means none of the data is stored on your Mac ever leaves the device. But if you’re still on the fence about the privacy aspect, you can separately disable indexing (and semantic search) for Messages and the files container.
From a functional perspective, Vector is extremely responsive, well-designed, and thoughtfully executed. The only area where it needs improvement is the semantic search and understanding. That’s something beyond the developer’s grasp. And it’s something that can either be fixed by fine-tuning the underlying AI model or using a smarter AI model.
For now, you can’t load an AI model of your choice. I would’ve loved to try one of Google’s Gemma series models, or those from the DeepSeek and Qwen families. Additionally, it would’ve been amazing to deploy specific AI models for certain tasks. For example, contextual image search would require a multi-modal AI for the best results.
There are already plenty of open-source models on Hugging Face that can pull it off. My experience with running SMoL-VLM2 on the iPhone 16 Pro for visual identification (even through the camera feed) was a fairly rewarding experience. Overall, if you’re looking for a low-stakes and minimally intrusive Spotlight alternative, Vector fills that gap pretty well. It’s only let down by the underlying AI brains in a few areas.
