Describe the process you want to automate: combine APIs, extract from PDFs and spreadsheets, or analyze exports with AI. Ipek builds the workflow on the canvas.
Record your screen while you talk the task through, or just talk. Speech is transcribed on your computer, and your AI reads the recording to learn the steps.
Claude Code, Codex or any other MCP client can operate Ipek from start to finish. You talk to your AI, and it plans with you in the Project document, then builds and refines the workers, workflows and dashboards.
Each worker is a small, deterministic program your AI writes: fetch from an API, read a PDF, merge, remove duplicates, analyze documents with AI, post to Linear. Workers take parameters, like an API key or a filter. You can open any worker and read its source code, which belongs to you.
Workers are wired together into workflows on a canvas. A workflow can branch, join, and run a step once for every file: for example, call two APIs, combine the data, and turn it into a PDF or CSV. Every run leaves its files in a folder on your computer, so you can check the results.
Each project has a built-in SQLite database for data a workflow needs to keep from one run to the next, like which tickets were already posted. Your AI sets it up and manages it, so there is nothing for you to write.
Once your workflows do what you need, ask your AI for a dashboard of the results: stock prices from your broker, feedback from your startup’s users, or analytics from World Bank Open Data. When others need to see it, publish it online.
Put a workflow on a schedule, like every 10 minutes, every weekday at 9:00, or once a month. Schedules run on your computer while Ipek is open.
A few of the automations built with Ipek.
Yeddi YüzApp feedback
Students flag questions in a quiz app. Ipek pulls the reports from PostHog, Claude checks each one against the curriculum notes and fixes the question files, and a dashboard shows what changed.
Market research
Every few minutes Ipek pulls new SEC filings, Claude scores each one, and the buy and sell calls go into a paper portfolio priced from a live broker feed. Each call is graded against the S&P 500 after 72 hours. No real orders are placed.
Both can automate the job. Here’s where each one lands for the work Ipek is built for.
Developer-focused tools like n8n, Windmill, and Pipedream can do a lot of this too — if you're happy running a server/docker or wiring up every step yourself. Ipek is the version for non-technical team leads who want to optimize their process: it installs like a normal app, and the AI plans & builds the workflows just for your needs.
Any MCP client, such as Claude Code, Codex or Cursor. Ipek has no AI of its own, so it adds no AI costs: you use the plan you already pay for.
No. Your AI writes the workers. You can open any of them and read exactly what it does.
Workflows run on your computer and recordings are transcribed there. What your AI reads goes to your AI provider, and a worker that calls an API sends what that API needs.
Ipek is free. Workflows run on your own computer, so there are no per-task fees.
No. Automations run while Ipek is open on your computer. Running them somewhere else is on the roadmap.
No account or payment needed. Everything runs on your own machine, offline.
Be the first to hear about new features, new releases, and Linux support as they land.