The new first run skips the Mac automation learning curve

Open a visual automation app for the first time and you get a blank canvas and a sidebar of sixty-odd nodes. Before anything useful happens you need to know what a trigger is, which node reads a PDF, how outlets connect, and why macOS is asking about your Downloads folder. Most people meet that learning curve on day one with nothing to show for it.

Our old first run leaned into the curve. It put you on the canvas and taught you to drag nodes. People who finished it could build a flow by hand. Plenty of people did not finish it.

The new one starts from a different fact: most Macs already have an AI on them, and an AI can build the flow for you. So the first run connects that AI first, has it build something you asked for, and shows you the result working before it asks you to learn anything. It is coming in the next release.

Find the AI you already have

Fig. 1Welcome: "Your Mac, on autopilot. What should we watch for today?" with node cards from a few flows drawn around it.

After Welcome, Watchflows scans the Mac. It looks for Apple Intelligence, the Claude Code and Codex command line tools, and a running Ollama or LM Studio server. It asks you nothing while it looks. Each row resolves to what it found, including a signed-out Claude Code or an Ollama with no models pulled, with the fix next to it.

Fig. 2Looking around this Mac: five rows, each showing what the scan found on this Mac.

Fig. 3Give Watchflows a brain: Your AI subscription, Run it on this Mac, Use an API key, and Not now.

Then you pick a brain. Your AI subscription uses the Claude or ChatGPT plan you already pay for, through its own CLI. Run it on this Mac uses Ollama or LM Studio, and waits for one to start if neither is running. Use an API key takes an Anthropic, OpenAI, Gemini or OpenRouter key and tests it with the provider before saving it. Not now skips ahead, and you build by hand.

Fig. 4Brain online: Watchflows asks one question and Claude answers in a line.

The next screen says "Let's see if Claude talks back." Watchflows asks it one short question and shows the answer. It takes a few seconds and settles whether the AI is really connected. If the answer does not come, the screen says why in plain words: sign in again, pick a smaller model, or re-enter the key.

There is no starter pack of our AI credits in here. The first run uses the AI you have. Cloud AI is something you can choose after you buy, not a meter running while you learn.

Build a first flow by describing it

Fig. 5Let's build your first flow: a text field to describe it, and four idea cards underneath.

Type what you want in a sentence, like "When a receipt lands in Downloads, rename it by vendor and date." Or start from one of four ideas: Rename receipts, Sort Downloads, Summarize PDFs into Notes or Sweep screenshots.

The flow then builds itself on the real canvas, the same one you will use tomorrow. A floating band narrates it. For a described flow it opens on "Consulting with Claude…" while the model works. Then it names each node as it lands and wires itself to the last one, like "Adding ‘Read the receipt’…" for the receipts idea. When it finishes, the band says "Built and wired." Watching a flow get made teaches the canvas better than a tutorial did.

Watch it work once, then turn it on

Fig. 6Try it after the run: each node shows what it did with the sample file, in and out.

Nothing is watching yet. Try it drops a real sample file into the folder the flow watches and runs it once by hand. Under each node, a short line shows what that node did with it: true on a check that passed, the new name on a rename, moved → Screenshots/Screenshot Watchflows.png on a move. The band ends on "It worked." and names what came out.

A flow you described gets the same treatment. Watchflows reads the flow its AI built and works out what can stand in for the trigger. A folder watch gets a file called Watchflows sample with the extension it expects. A schedule gets a tick, a clipboard trigger gets a copy of "Hello from Watchflows". If nothing can stand in, such as a calendar event or a device being plugged in, it says so before running anything and offers Turn it on in place of Try it. A described flow never fails its first run because the sample was wrong.

The folder is where macOS steps in. Touching Downloads for the first time raises the system prompt asking whether Watchflows may use that folder. So before the drop, the screen says so in plain words and tells you to choose Allow. If you choose Don't Allow, the flow is saved and left off, with a button to the right pane in System Settings.

Fig. 7Turn it on: the band reads "It's live." and the flow is watching Downloads.

Turn it on arms the trigger. From then on it runs by itself, even with the window closed.

Decide, then you're set up

Fig. 8Make it yours: Cloud, Cloud Pro, Own it at $49, or keep the 14-day trial.

Make it yours comes last, after the app has already done something for you. Pick Cloud, Cloud Pro, or Own it for $49 once. Or keep the 14-day trial, which charges nothing automatically and reminds you three days before it ends.

Fig. 9You're set up: all four moments reached, each with what it was.

You're set up shows four moments, which are how we now measure whether a first run worked: Brain online, Watched it work, Live and Decided. Each one is a thing you saw happen, not a screen you clicked past.

Already using Watchflows?

Choose Help ▸ Show Welcome and the whole first run plays again, any time. Your flows stay exactly as they are; what you build during the replay is added next to them. New installs get the first run on first launch.

It is also the quick way to connect a brain you added since. Install Ollama or sign in to Claude Code, then run Show Welcome and let the scan find it.

Try a description of your own

The describe field takes a sentence like this one. Paste it in on the first run, or later into the Flow Builder.

Paste into the Flow Builder

When a PDF lands in Downloads, summarize it in five bullets and save the summary to a new note in Apple Notes titled with the file name.