Designing an app · Lesson 1
A manifest is a file that describes an app: who uses it, which data it keeps, which pages it shows, which agents work, what happens by itself, where a person must decide, how people pay. It contains no code: one runtime reads it and becomes that app. Here we design a small one, ContractCheck, by answering nine questions in order, each one adding a few entities to a graph.
Use Next, or the ← → keys. At each step the platform’s real validator checks the manifest in your browser; try the typical mistake to see the rule it enforces.
ContractCheck uses the most common entities. The others follow the same logic:
data.jobs — ETL pipelines on the platform: an existing one, one written in the manifest, or one an agent builds and runs; their rows can land in a collection or a search index.data.credentials — API keys users connect on their account page, encrypted, used by http tools.schedules — App-wide recurring work: “every day at 5 the monitor checks every account”, with the report emailed to a role.templates — Personal agents each user creates for themselves, with their own parameters, schedule and data.teams — Several agents published as one: a sequence of steps (a DAG) or a swarm with a lead and a shared task board.slots — Small Python functions run in a WebAssembly sandbox, without network or files, for deterministic calculations.channels · serviceHours — Phone numbers, Telegram bots and mailboxes connected to an agent; the hours when it must answer fast.mobile — The Android and iOS app: one shell that loads the same runtime, with only the phone capabilities declared.limits · exports · notifications — Rate limits and quotas per plan; endpoints in JSON, CSV, XML or RSS; an email when a report is ready.From here the path goes into each area in detail, starting from the one every safety rule depends on: the data, and whose each record is.