← Back to Concepts

Concept · Set of pages

Submit, poll, read: what a long analysis looks like over HTTP

An operation on this platform does not return a result.

Page 1 of 4

By Ruben Achterberg · Staff engineer, application runtime · 5 August 2026 · 8 min read

The first thing to internalise about calling an application here is that a call does not wait. You submit an operation, the application accepts it and answers immediately with a job id, and the work happens in a background worker process that is not the one holding your connection.

That is not a limitation being apologised for. An analysis that reconciles nine ledgers or scores twelve hundred routes takes minutes, and a minutes-long HTTP request is a request that dies to a proxy timeout somewhere in the middle, at which point nobody — including the caller — knows whether the work happened.

The three requests

Every operation on every application follows the same shape, which is the point of having a common engine.

  • GET /operations — what this application can be asked, and the input each operation expects. This is discovery: you do not need a client library to find out what an application does.
  • POST /ops/<key> — ask it. The response is 202 with a job id in it.
  • GET /jobs/<id> — how it is going. Poll this until the status is terminal.
  • GET /outputs — what it produced. The result is a file in your organisation's own store, not a blob in the polling response.

Used in