WORKED EXAMPLES
Build something that keeps working¶
A worked example is a complete agent, built end to end: the plans it runs, the permissions it holds, and the receipts it leaves behind.
This page is for whoever has finished the quickstart and wants to see the whole thing rather than one step of it.
The quickstart gets you a governed run. A worked example gets you an agent that is still doing useful work next Tuesday, without you.
Every example here is a real transcript, captured from a run rather than written out by hand.
A different meaning of the same word
Elsewhere in EKKA a recipe is a specific thing: one program on one machine, with a fixed argument list, that a plan may run. These pages are worked examples and have nothing to do with that.
Each example assumes you finished Build a governed AI application: you have the CLI, an organization, and an Enclave running on a machine you control.
-
A folder on your own disk that is the agent's memory, a plan that adds to it on a schedule, and a weekly memo written by a model using a prompt EKKA never reads. Start here if you want to see what "your data, your machine" means in commands.
What does every example have in common?¶
- Your infrastructure. The Enclave runs on your machine. Your files, your credentials and your model output never pass through EKKA.
- Declared permissions. The agent lists what it wants before it runs, and a plan asking for anything outside that list is refused when you save it.
- A receipt per step. Signed, hash-linked, and checkable offline with
ekka receipts verify. - Stated limits. Each example ends with a Limits section: what is true about the path it just taught.