Jeeves - Turn repeat work into reusable AI workflows
by•
Jeeves is a desktop app for reusable AI workflows. Set up the steps for a task, run them with your input, and review the result. Start with a weekly update: notes in, draft and check, update out. Change the steps in a visual editor and reuse them next week. Add human input when the workflow should wait for you. Free and open source for Mac, Windows, and Linux.
Replies
Pictioner
Hi, I'm Priyankar. I built Jeeves for work that comes around again: the weekly update, the research brief, the next batch of requests.
Take a weekly update. Add your project notes, have an AI draft the update, check it against the notes, and review the result. Save those steps. Next week, give the same workflow new notes.
You can start with a template and change the steps in a visual editor. Each step has a job, such as researching, writing, or using the browser. You can see its inputs and results. If it needs a login, an answer, or your review, Jeeves saves progress and waits for you to continue.
Under the hood, you choose the AI service for each step. Jeeves supports OpenAI, Codex CLI, OpenRouter, and local models. Optional Jev decisions can choose the next step when a workflow needs to branch. Human input steps save progress and wait for your answer.
For a closer look at decisions and human input, the technical walkthrough is here:
Start with the built-in example, which uses sample data and needs no API key. For your own tasks, connect an AI service. The app is free under Apache 2.0, with no paid tier. Your chosen provider may charge for usage; live Jev decisions need a TypeSafe key.
The name comes from Wodehouse's Jeeves, the person who knows what needs doing. A small nod to a very capable assistant.
This is an early preview for Mac, Windows, and Linux. Mac downloads are signed and notarized. Windows is unsigned, updates are manual, and browser tasks sometimes need your help. Your workspace stays on your computer; cloud AI services receive the task context you send them.
I'd love to hear which recurring task you'd try first, and where setup gets confusing.