I Retired Raycast

Raycast was one of the first apps I installed on every new Mac, and I used it constantly for years. AI made credible alternatives easier to build while giving Raycast strong incentives to become a larger AI product I no longer needed.

Raycast is a Mac app launcher. You hit a keyboard shortcut, type the first few characters of an application's name, select it, and launch it. If you like to keep your hands on the keyboard, it's great.

I've been using Raycast for years. It has continued to evolve, adding extensions, calendar controls, Spotify integrations, clipboard history, window management, and plenty of features I never explored. More recently, AI became a substantial part of the product. You can ask questions, run commands against selected text, and give an agent access to extensions that act on other applications.

The app has served my needs well for a long time, but it was also growing in a direction I didn't need.

I saw someone on X mention recently that they were uninstalling Raycast and moving to Tinycast, a free, open-source Mac launcher. That was pretty much all it took for me. Smaller, faster, and open source.

It also gives me AI features without asking me to buy another pool of AI usage. Tinycast can use the Codex, Claude, or OpenCode command-line tools already installed on a Mac, with the accounts already signed in there. I already pay for more than one of those subscriptions and usually have AI sessions running in other terminal windows.

My launcher doesn't need to sell me more intelligence. It needs to connect me to the intelligence I've already purchased.

AI has a real marginal cost

This isn't a complaint that Raycast should provide expensive AI features for free.

Model calls cost money. That cost becomes especially difficult under a flat subscription because the users who get the most value from a product may also generate the highest bill. At the same time, I don't want to buy AI from Raycast. I'm already going direct.

Raycast now lets paid subscribers connect Claude or ChatGPT, but access still starts with a paid Raycast plan.

That is a reasonable product decision. It just doesn't make sense for me.

I don't need another AI interface enough to pay for Raycast Pro. Tinycast gives me the thinner layer I want for free.

AI adoption is splitting in two

Raycast and Notion have a clear reason to put AI inside products people already use. Every useful AI feature gives an existing customer one less reason to leave for an AI-first competitor or a product from one of the frontier labs.

I don't know how long that defense lasts. The buying decision depends on which layer a customer thinks is replaceable.

For many people, the application is the product and the model behind its AI features is interchangeable. Buying software with AI already included gives them a supported path into the technology. They can just use the product they already understand and discover an AI feature inside it.

Today I spoke with a CEO and his team as they started experimenting with agents in Microsoft Copilot. The integrated experience made it easy to summarize a document or ask questions of the HR handbook. Once they wanted to analyze a financial model, we had to start thinking about which model was doing the work.

At the other end are people whose AI usage already cuts across their tools. We have subscriptions, local models, command-line agents, and our own opinions about which system should handle a task. An application that bundles another AI layer can feel less like convenience and more like duplication.

I already have the model subscriptions I want. The launcher that gives me access to them is replaceable. Tinycast won because it let me keep those subscriptions and swap out the interface around them.

Those two groups place a different value on the same integration. One wants the software company to make all the choices. The other wants the software to stay out of the way and connect to choices already made elsewhere.

The ecosystem came with me

The subscription is only half of why switching was so easy. Tinycast also brought much of Raycast's ecosystem with it.

Raycast didn't only build a launcher. It built an ecosystem of extensions that made the launcher useful for thousands of specific jobs. An extension might search a service, control an application, or expose a workflow through the same keyboard interface. That catalog should make Raycast harder to replace.

I have to imagine that AI-aided software development made it easier than ever for Tinycast to adopt parts of the Raycast ecosystem and replace Raycast on my Mac.

This is a strange consequence of software becoming easier to build. A smaller product can focus on a narrower experience, connect to AI capabilities that already exist elsewhere, and implement compatibility with the incumbent's extensions. It doesn't need to reproduce every feature or persuade every user. It only needs to serve enough people who want a thinner version of the product.

That doesn't mean Raycast has no moat. Its polish, reliability, product judgment, distribution, and continued investment all matter. Most people won't inspect an open-source launcher, install it through Homebrew, or accept the maintenance risk of a smaller project. Plenty of users will pay to have Raycast make those concerns disappear.

But the boundaries around the product are more porous than they used to be. The intelligence comes from model providers. The extensions can run somewhere else. The replacement is easier to build. Users who are comfortable assembling their own stack can leave behind the parts they no longer need.

I retired Raycast because Tinycast does the smaller job I was hiring it to do. Someone else may make the opposite choice because they want the larger experience Raycast is becoming.

Both decisions make sense. Even as steeped in this shift as I am, I keep getting surprised by how quickly the economics change beneath software I use every day.

Sources

• Tinycast
• Tinycast AI Chat documentation
• Tinycast extension documentation
• Raycast on its new AI pricing