When I receive a document that I might need later, I scan it, store it in Google Drive, and throw the paper away.

Bank letters, medical documents, invoices, administrative forms, random contracts… all of it goes through the same process.

If you’ve ever scanned a document with your phone or a scanner, you know the result. You end up with files called:

Scan_0037.pdf
Document.pdf
IMG_8421.pdf

Completely useless names.

When you accumulate hundreds of documents over the years, this becomes painful. Even if the content is searchable thanks to OCR, filenames still matter a lot. They are the first thing you see when you browse folders.

And strangely, even companies like Google, who are leaders in AI, still haven’t solved this simple problem in a satisfying way inside their products.

Recently I’ve seen a lot of small utilities appear that promise exactly this. They use AI to read a document and generate a clean filename.

The idea is good. The price less so. Some of these apps charge around $10 per month just to rename files.

That’s the kind of moment where I usually pause and think: there must be a simpler way. And it turns out there is.

Using Apple Shortcuts and Apple Intelligence, you can build this workflow yourself in a few minutes. It runs directly on your Mac, can work entirely locally, and it’s reliable.

Let me explain how it works.

The idea behind the workflow

The logic is very simple.

When you select a PDF, the shortcut does four things:

1. Extract the text from the document

2. Send that text to an AI model

3. Ask the model to generate a structured filename

4. Rename the file automatically

That’s it.

The only real trick is defining a clear naming format so the AI behaves consistently.

After some experimentation I settled on something like this:

YYYY MM Sender Topic

For example:

2024 11 HMRC Tax Letter
2023 09 EDF Electricity Bill
2024 02 NHS Blood Test Results

This format works well for several reasons.

Putting the date first keeps documents naturally sorted.
Adding the sender makes folders much easier to scan visually.
And the topic gives just enough context without making filenames too long.

The goal isn’t to describe the document perfectly. It’s simply to make it instantly understandable.

Building the Shortcut

Everything happens inside the Shortcuts app on macOS.

You don’t need any external software.

Step 1 : Receive the file

Open the Shortcuts app on your Mac.

Then click the “+” button to create a shortcut.

Make sure the shortcut is configured to receive files in Finder, and that Services Menu is enabled so the shortcut appears when you right-click a file.

Once saved, the shortcut will be available from Finder through:

Right click on a file → Services → your shortcut.

This allows the shortcut to run when you right-click a file in Finder and trigger it from the Services menu.

So the workflow becomes:

Right click → Services → Rename with AI.

Select : Receive Files from Quick actions.

Step 2 : Extract the text

Add the action:

Extract Text from File

This performs OCR on the document and produces raw text that the AI model can analyse.

Without this step, the AI wouldn’t know what’s inside the document.

Step 3 : Ask the model to generate the filename

This is where Apple Intelligence (or ChatGPT) comes in.

You add a model action and give it a very strict prompt so it behaves like a classifier rather than a chat assistant.

For example:

You are a highly precise document classifier. Read the following text extracted from a scanned document. Your ONLY task is to generate a highly qualitative file name.

The format must strictly be: YYYY MM Sender Topic

Rules:

YYYY MM is the date mentioned IN the document (not today’s date). If no exact date is found, use the closest approximation. If you can’t find an approximation, then use today’s date.

Sender is the company, person, or institution who issued the document. Use space between each word.

Topic is 2 to 6 words max summarizing the document (e.g., Invoice, BloodTest, RentReceipt). Use space between each word.

Do NOT output the .pdf extension.

Do NOT output any other text, no introduction, no punctuation at the end. ONLY the file name.

Use the same language for the filename as the docuemnt. Is the document is in french, the filename should be in french. If the document is in english, the filename should be in english.

The stricter the prompt, the more reliable the result.

Step 4 : Rename the file

Finally you add:

Rename File

And pass the AI response as the new filename.

At this point the shortcut already works.

Select a PDF, run the shortcut, and the document gets renamed automatically.

On-device vs Private Cloud Compute

One interesting detail inside Shortcuts is that you can choose which Apple model to use.

You have 3 options.

1. On-device model

This runs entirely on your Mac.

Advantages:

• very fast

• no data leaves your machine

• works offline

For many documents this is already good enough, especially if the scan quality is decent.

2. Private Cloud Compute

Apple also provides a privacy-focused cloud model.

This can sometimes produce better results when:

• the document is messy

• the OCR text is noisy

• the structure of the document is unclear

So the trade-off is simple.

On-device is faster and fully local.

Private Cloud Compute can be slightly more capable for complicated cases.

Both work well for this workflow.

3. ChatGPT

Shortcuts also allows you to use ChatGPT instead of Apple’s models.

I experimented with this as well.

ChatGPT sometimes produces slightly better filenames when the document is messy or ambiguous. In some cases it also feels a bit faster.

So in practice you have three possible setups:

1. On-device Apple model

2. Apple Private Cloud Compute

3. ChatGPT

They all work well.

The choice mostly depends on whether you prioritize privacy, reliability on difficult documents, or raw output quality.

Handling multiple files at once

The first version of the shortcut only works with one file.

That’s already useful, but the real improvement comes when you add a loop.

Inside Shortcuts you can wrap the workflow in:

Repeat with Each

The logic becomes:

Receive selected files
Repeat with each file
Extract text
Generate filename
Rename file

Once you do this, you can select ten scanned documents at once, run the shortcut, and watch them get renamed automatically.

That’s where it starts becoming really powerful.

Integrating the workflow with Raycast

There is one final improvement that made the workflow even better for me.

Instead of triggering the shortcut from Finder, I run it directly from Raycast.

Raycast can launch macOS shortcuts natively.

But there is one small change needed.

Instead of starting the shortcut with:

Receive Files from Quick Actions

you replace it with:

Get Selected Files in Finder

This allows the shortcut to detect which files are currently selected in Finder when Raycast launches it.

So the workflow becomes extremely simple.

• Select files in Finder

• Open Raycast

• Run the shortcut

And the files are renamed instantly.

For someone who already uses Raycast as a command centre for the Mac, this feels much more natural than right-click menus.

Why I find this interesting

What I like about this setup is not just that it works, it’s that it shows something broader.

We’re entering a moment where small AI-powered utilities are appearing everywhere, often charging subscriptions for very narrow features.

Sometimes those tools are useful. But sometimes the underlying capability already exists in the operating system.

In this case, macOS already gives you:

• automation

• OCR

• local AI models

• cloud AI models

• file system integration

All the ingredients are already there, you just have to connect them.

This also works for other things

Once you understand the pattern, you can apply it to many other tasks.

For example:

Automatically rename images with SEO-friendly names
Classify documents by category
Clean up messy archives

The core logic is always the same.

Extract information from the file, ask a model to interpret it, and apply a structured action.

Renaming documents is simply the easiest place to start.

About a year ago, I switched to Raycast.

This is not a criticism of Alfred. I still have a lot of respect for what the team built. In many ways Alfred introduced me to the entire concept of launchers. I know tools like Quicksilver existed before, but Alfred is what made that world accessible to me.

This is simply a personal reflection on why Raycast now fits my workflow better.

What is Raycast

Raycast is a productivity launcher for macOS and Windows.

It lets you launch applications, search for files, or quickly open browser bookmarks without leaving the keyboard. Very quickly though, it becomes much more than that.

Raycast also includes small built-in tools that tend to become indispensable over time, like a clipboard history or snippets (Snippets are small pieces of text you can insert anywhere using a shortcut, which is very useful for repetitive writing).

Once you go a bit deeper, Raycast starts revealing more capabilities aimed at advanced users. You can create script commands, which allow you to run custom scripts directly from the launcher, or install extensions that add entirely new features.

These extensions can either be built privately for your own needs, or installed from the Raycast Store, where thousands of community extensions connect Raycast to different services, APIs, and workflows. Over time, this ecosystem turns Raycast from a simple launcher into something closer to a personal command centre for your computer.

The Alfred workflow ecosystem problem

One of the main friction points for me with Alfred was the ecosystem around workflows.

Originally, workflows were shared all over the place. Some were on forums, some on random websites. You often had to search manually to discover useful workflows. Installation and updates could also be inconsistent depending on where the workflow came from.

A store eventually appeared, the Alfred Gallery, which definitely improved things. But compared to what exists today in Raycast, it still feels fragmented.

Raycast approached this differently. Extensions live in a centralised store, with a clear installation flow, discoverability, versioning, and maintenance. There is also a team behind the scenes reviewing and maintaining the ecosystem, which improves quality and reliability.

For me this makes a big difference.

Where Raycast really changed things for me

The biggest reason I stayed with Raycast is the extension system. Not just the public extensions, but the ability to build private ones.

I started creating small internal tools connected to APIs from my own business systems. And that significantly changed how I work.

For example, I have a weekly banking reconciliation process for my business. Before, I had to open my CRM, find each invoice, and manually mark them as paid via transfer or direct debit.

Now I have a Raycast command that connects directly to the CRM API. I can search an invoice from Raycast and mark it as paid in seconds using keyboard shortcuts.

The difference in friction is huge.

Automations that live inside the launcher

Once you start going further with script commands and private extensions, that you can create with the help of tools like Claude Code or Codex, you can build extremely powerful workflows.

More broadly, we are entering an era where companies can use modern coding tools to create highly customised software experiences tailored to their exact internal workflows, which feels like a meaningful shift in how software is built and used.

For example I built a small system for unpaid invoices.

With one shortcut in Raycast I can:

• list all invoices unpaid after X days
• review the situation
• generate a reminder email for the customers, with all the details

All without leaving the launcher.

These are very niche automations tailored to my own workflows, but that is exactly the point. Raycast makes it surprisingly easy to build these kinds of tools.

The free tier is surprisingly generous

Even without the Pro subscription, Raycast provides a huge amount of value.

The main paid features revolve around AI. But the team made an interesting decision: you can bring your own API key.

If you connect your own OpenAI or other LLM API key, you can access almost everything the AI features offer without paying for the Pro subscription.

The trade-off, of course, is that if you start making a lot of requests to expensive models, your API usage might end up costing more than the subscription itself. But at least the option exists, which I find quite fair.

The only real limitation in free version that remains is sync across multiple machines.

Venture capital and the Raycast debate

A topic that often comes up in discussions about Raycast is venture capital.

Some people are worried because Raycast is VC-funded and therefore under financial pressure to monetise.

I understand the concern. But at the same time, venture capital is also what allows ambitious products to exist in the first place.

Raycast is not just a technical launcher. There is a whole team behind the product: people designing the extension system, maintaining community extensions, reviewing submissions, improving UX, and pushing the product forward.

That kind of product development requires resources. And what Raycast offers for free today is already quite remarkable.

The hype

One thing I sometimes find slightly frustrating with Raycast is the marketing.

Their “hype team” does an incredible job building excitement around the product. Maybe a bit too good.

Sometimes announcements come very early, which creates long periods of expectation. For example, they teased the next Raycast Mac update several months ago with a video, and people have been waiting ever since.

When you build that kind of anticipation, delays can easily turn into disappointment.

Personally I also feel that, at times, the tone on social media can become a bit exaggerated. The goal is often to create excitement and a sense that something big is coming, which naturally generates a bit of FOMO in the community.

Personally I prefer when companies announce things closer to release, in a more subtle way.

But I can I also recognise that building a narrative around a product is important. Vision matters. It just sometimes feels a little over-hyped

Thoughts about Glaze

Raycast recently announced Glaze, which allows people to build Mac applications.

I find it interesting, but I also feel slightly conflicted. On one hand, I love seeing them innovate. On the other hand, Raycast originally had a very clear philosophy: avoid traditional apps and bring everything into a single fast interface controlled by the keyboard.

Glaze seems to move back toward separate native applications with their own interfaces.

Personally I use very few traditional apps, so I am curious to see how this evolves. Here is my current dock :

Part of me wonders if it might have been interesting to expand the UI capabilities of extensions instead. Right now extensions have limitations in how they can display complex interfaces.

If extensions could support richer UI, we could build much more powerful tools while still staying inside the Raycast interface.

Maybe Glaze will complement that vision. Time will tell.

Final thoughts

Again, this is not about Alfred vs Raycast.

Alfred remains an incredible tool and it played a huge role in how many of us discovered launchers in the first place. For my current workflow though, Raycast fits better.

The extension ecosystem, the developer experience, and the ability to build custom internal tools directly inside the launcher have made it an essential part of how I work every day.

In a future article, I’ll probably walk through my full workflow for building private Raycast extensions for my own business, from start to finish. If that sounds interesting, feel free to subscribe so you don’t miss it.

I’m very curious to see where the Raycast project goes next, and I’m excited to see what the team builds in the coming years. They’ve already created something remarkable. It’s not easy to design a tool that becomes such a central part of people’s daily workflows, and the Raycast team clearly deserves a lot of credit for that.