Product · Configured calls
Configure the call. Write the logic.
A call to another system is configured once — address, credential, retries, paging, what counts as an answer — and named in your words. The script calls the name. Everything that is not logic has a place of its own, and each place is tested before a script exists.
HTTP: configured in a minute, tested at once
Method, path, headers, a body with named values. Press Test and you see the answer your script will get and the request that was sent — the credential named, never shown. A GraphQL API is called the same way: a POST whose body carries the query and its variables.
- The answer is a decision: which statuses are an answer, and what part of the body the script gets. Anything else stops the run with the far end's own status and body.
- Transient failures are repeated in place, 3 times by default; 400, 401, 404, 422 are never repeated. Name a status as a long wait — 429 is the usual one — and the whole run is parked for the time the far end asked, up to 24 hours, holding nothing open.
- Paging is configuration: next-link, cursor or offset. One line reads every page into a dataset and keeps its place across a throttle.
Text and email: the wording belongs to the person who owns it
A comment on a ticket, a message to a customer, a mail with its subject, body and attachments: each is a template with named values, kept outside the code. The person who owns the wording — a support or operations engineer, not a developer — changes it, tests it with a sample and activates it. The script keeps calling the same name.
The function types in the docs →
Mappings: the CSV-and-JSON fiddling leaves the code
One JSON shape into another, or CSV into records and back, declared as a template. Type a sample and the result is shown as you type — the same mapping a run uses, so what the preview shows is what the script gets. A change to a mapping is a configuration change.
JSON and CSV mappings in the docs →
Model calls: everything about the call in one place
The prompt, the model, the connection that holds the key. The call goes through the same path as every other call, so it is recorded, repeated and paused the same way. The providers are anthropic and echo: the second answers deterministically without a key, for building and testing. A model's answer is input, and the script validates it like any other.
The model call type in the docs →Your names complete in the editor
Every configured call completes and hovers in the script editor with its signature and what it answers — before anything is activated. A name that resolves to nothing is marked while you type.
Writing scripts in the docs →
What it will not do
- A call goes to its connection's address only; a path or continuation link that would leave it is refused.
- A bot-check page answering instead of the API is detected and named, never solved.
- A credential is never a value a script, a log, a diff or a recorded request can show.