Hand over a job. It learns how you do it.

Claritty builds and runs automations for the work you repeat by hand. You describe the job in plain language, Claritty wires it into the tools you already use, and it runs on a schedule. You see what it did every time, and anything that acts on your behalf waits for your approval. Start from a ready-made automation or describe your own.

ChatGPT helps you do the job faster. You still do it. A Claritty automation does the job itself, on its own schedule, against your real tools, whether or not you open anything. An app builder hands you code and leaves you owning its maintenance and its security. Here the automation runs on the platform and your credentials are brokered server-side, so they never reach app code.

With Zapier or Make you assemble the steps: pick a trigger, add the actions, map the fields between them. That suits a rule you can state exactly. With Claritty you describe the outcome and it composes the automation, then puts an agent on the steps that need a decision rather than a rule. The graph is something it produces, not something you maintain.

Three ways, and all of them stick. Correct a piece of its work and what you said becomes a standing rule. You can see which agent it binds, what you said that produced it, and turn it off. Answer a question it asks and it keeps your answer instead of asking again. Or show it: do the job once in your browser and say why as you go: the clicks become a job it can run, and what you said becomes rules it keeps. New rules go on trial for five runs before they settle, and nothing takes effect because a screen was looked at.

Then you show it. Install the Chrome extension, do the job once the way you always do, and read back the steps it wrote. It watches what you do rather than the data on the page, and it asks when a step could mean two things. That is the work no connector reaches: a carrier portal, a council system, the back office you log into every Monday.

Every action leaves a receipt: what was done, in which system, and the reference that system gave back. Where the other side returned no reference it says so rather than showing you a tick it cannot support. Anything you did yourself is recorded as your word with no receipt from anywhere, because that is what it is.

Only where you have said it can. Every write waits for a person, and the message you get carries what it read, what it is about to do, and the exact words it will send. You can approve that one run, ask for changes, discard it, or approve and stop being asked for that job.

Safety is architectural, not a promise: integration credentials are AES-256 encrypted with AWS KMS and brokered server-side, so they never touch app code. Each app runs in its own isolated runtime with its own database schema. See our Security page for the full control-by-control picture.

Free is $0, no card: build up to three automations and actually run them, five times, taking real actions on a schedule. Pro is $25 a month and lifts the five-run ceiling, adds building your own agents, 2,000,000 credits and a smarter model. Business is $129 a month with 12,000,000 credits, your whole team, the most capable model, and bring-your-own-key. Credits refill every cycle.

No. Ready-made automations install in one click with a guided setup, and building your own is a conversation rather than code. Developers who want full control can also build with Claude Code or Codex and publish via the Claritty CLI.