WorkerKit

AI workers for the job you need done

Browse ready-made AI worker kits by the job they do. Pick a job, connect the apps it needs, and the worker is on it in about a minute.

What jobs can an AI worker do?

The ones that repeat and follow rules: triaging an inbox, chasing a reply that never came, keeping records straight, turning a week of activity into a report. A kit packages one such job with the instruction, the app permissions, the memories and the schedule it needs, so the work runs unattended instead of waiting for someone to get to it.

How a job becomes an AI worker

Every kit here is written for one job, and it ships with the four things that job needs in order to run without you.

The instruction is the job written out: what to do, in what order, and what to leave alone. The publisher wrote and tested it, so there is no prompt to engineer.

The app permissions are the accounts the job touches and the level it gets on each one: Off, Read or Write. You approve every connection yourself, and a worker can never reach an app you have not connected.

The memories are the facts and rules particular to you, in your own words: who your team are, how your company phrases things, what counts as urgent.

The schedule is when the job runs: a wall-clock time in your own time zone, or on demand when you ask. Every run leaves a receipt showing what the worker did.

Pick the job below and the kit pages under it state all four before you install anything.

Questions about picking a job

Can one worker do more than one job?

A worker is one job by design, because the instruction, the access it needs and the schedule it runs on all belong to that job. Two jobs means two workers, which is also what lets you give each one only the access its own job requires. Every plan runs several workers at once.

What if no kit does my job?

Build it. Describe the job to the AI assistant you already use, connected to the WorkerKit MCP server, and it writes the kit for you: the instruction, the apps and the access each one gets. It validates the kit and runs it once before anything ships. You can keep the kit private to your account or publish it here, where it gets a page and an address for life.

How long does it take to get one running?

About a minute for a kit that uses accounts you have already connected: pick the kit, approve the access it asks for, choose a model and deploy. The part that usually takes longer, writing the instruction and the access rules, was done by the publisher.

Which AI model does the worker use?

Claude, ChatGPT, Gemini and Grok can run language model kits. You can switch a worker to a different model or send a single run to another one to compare them on the same job. Per-model grades are not currently displayed on kit pages; compare results on your own examples.