AI features

Echo includes a set of AI assists for the moments where an extra pair of eyes saves real time. They are optional, seat-based, and designed so your secrets never leave your machine.

What the assists do

  • Explain any response or error. Select a confusing payload or a failing call and get a plain-language explanation.
  • Generate a request from a pasted spec. Paste the relevant part of an API description and Echo drafts the request for you.
  • cURL import assist. For cURL commands too mangled for a plain import, the assist helps reconstruct the intended request.
  • Debug errors. Work through a failing request with guidance instead of guesswork.
  • Generate assertions and full test suites. Turn a working request into checks the collection runner can execute, from a single assertion to a whole suite.
  • Generate mock servers from a schema. Hand over a schema and get a local mock to develop against.
  • Document flows. Produce documentation for a sequence of requests, useful before publishing docs as described in /docs/echo/import-export.
  • Optimize request chains. Let the assist review a chain and suggest a tighter version.

You need a Pro AI seat

AI features require a Pro AI seat. Seats are managed by your organization on the Gapfy platform, so if the assists are not visible in your Echo, ask your organization admin whether a Pro AI seat is assigned to you.

Everything else in Echo, the five protocols, collections, environments, automation, import and export, works without AI and without a Pro AI seat. The assists are an addition, not a dependency.

Your secrets stay yours

Before any AI request leaves your machine, Echo redacts secrets deterministically. Tokens, keys and other sensitive values are stripped out locally, every time, by rule rather than by judgment call. The assist sees the shape of your problem, not your credentials.

This fits Echo's local-first design: your database is encrypted at rest, guest mode works fully offline, and AI is a feature you opt into with a seat, not a background process.

Getting the most out of the assists

  1. Start with explain-response on an API you are learning. It is the fastest way to build a mental model.
  2. When a request works, generate assertions for it, then let the collection runner guard it from regressions. See /docs/echo/automation.
  3. Use mock generation from a schema when the dependency you need is not deployed yet.