Everything FlowRunner can do.

On-Device AI Step — local LLM workflow, no internet

The AI Step is the heart of FlowRunner. It sends text from your workflow to a large language model — such as Llama 3.2 — running directly on your Mac, then passes the model's response to the next step. Because the model lives on your machine, nothing is uploaded to a server and the step works with the network completely off.

You write the instructions in plain English, so a single AI Step can summarize a long document, extract structured data from messy notes, rewrite content in a different tone, classify items into categories, or generate a formatted report. The same step adapts to dozens of tasks just by changing what you ask it to do, which makes on-device AI automation on a Mac feel less like programming and more like delegating.

Read Files Step — bring local files into the flow

The Read Files step pulls content from your local disk into the workflow. It reads plain text, Markdown, source code, and PDFs, extracting the text so downstream steps can work with it. You can point it at a single file or use glob patterns to match many files at once — for example every .md note in a folder.

Everything the step reads stays local. It simply makes your files available to the rest of the flow, whether that is an AI Step that summarizes them or a Transform that cleans them up first.

Transform Step — clean data without AI

Not every task needs a language model. The Transform step filters, sorts, deduplicates, combines, and reformats data using deterministic rules — no AI, no guessing, and no tokens. It is fast, predictable, and ideal for preparing a dataset before an AI Step or tidying the results afterward.

Use it to drop empty rows, sort entries by date, remove duplicates from a list, or merge several inputs into one structure. Pairing a Transform with an AI Step often produces cleaner, more reliable output than either could alone.

Template Step — structured output, one per item

The Template step turns raw data into nicely structured output by injecting values into a layout you define. It can repeat per item, so a workflow that processed ten documents can emit one cleanly formatted summary per document. Output can be Markdown, HTML, or plain text.

This is what makes FlowRunner's results feel finished rather than raw. Instead of a wall of model text, you get consistent, repeatable documents shaped exactly the way you want them.

Write File Step — save results locally

The Write File step saves a workflow's output to your Mac as a .txt, .md, or .csv file. You can write a single combined file or one file per item — for instance a separate Markdown summary for each note in a folder.

Because the output lands on your local file system, your processed data is yours immediately: searchable in Finder, editable in any app, and never parked on someone else's server.

Run Command Step — extend the flow with shell commands

Available in Pro, the Run Command step executes a shell command as part of your workflow, opening the door to advanced integrations — calling a script, converting a file, or kicking off another tool. Before any command runs, FlowRunner shows an explicit safety prompt so a command never executes without your direct approval.

This step is what lets technical users wire FlowRunner into a larger local pipeline while keeping the on-device, privacy-first model intact.

AI Workflow Builder — describe it in plain English

You don't have to assemble steps by hand. With the AI Workflow Builder you describe what you want in plain English, and FlowRunner builds the steps for you. Then you refine the result by chatting — adjust an instruction, add a step, change the output format — and the assistant updates the workflow.

The builder even checks its own work, validating that the steps connect sensibly before you run them. It is the fastest path from idea to a working no-code AI automation on your Mac.

Model Selection — tiny and fast to large and precise

FlowRunner lets you choose the local model that fits your hardware and your task. Tiny models are fast and use very little RAM, which is great for quick, high-volume jobs. Larger models trade speed for accuracy on complex reasoning. Each model is downloaded once and then runs locally forever after, fully offline.

Optional Cloud Models — bring your own key, strictly opt-in

If you want the quality of a frontier cloud model for a particular step, you can bring your own Anthropic or OpenAI API key. The key is stored securely in the macOS Keychain and is never transmitted to FlowRunner. Cloud AI is off by default and entirely opt-in — when you do use it, the text for that step goes directly from your Mac to the provider you chose, not through any FlowRunner server.

Live Run Timeline — watch every step happen

When you run a workflow, the Live Run Timeline shows each step's status in real time, so you can see exactly what is happening as it happens. You can stop a run at any moment, and when it finishes, clickable links take you straight to the output files on your Mac.

Curious how the steps fit together end to end? The FAQ answers common questions about models, privacy, and pricing.

Private AI automation, free to start.

FlowRunner is a one-time $9.99 Pro unlock on the Mac App Store, with a free tier that runs entirely on your Mac. No account, no subscription, no data leaving your device.