Documentation

What Sapilon is

Sapilon turns a legacy system or an AI-built prototype into production-ready, AWS-native software you own outright, using a governed AI agent whose permissions are written into the code itself.

Your first project

Create a project from the projects list, give it your existing code or a written description as input, and open it — each project is an isolated workspace with its own repository, wallet draw, and stage.

The four stages

Every project moves through Design (shape, mocked data), Build (real backend on the Build Server), Rehearsal (dress rehearsal on the Rehearsal Server), and Live (production) — and each transition is a gate that checks readiness rather than a deploy button.

Your first Ask

Ask for one outcome at a time, say what it is for and what must not change, and read the diff before you accept it — a good brief is the difference between one turn and five.

Existing System

Answer one question about where your current system lives, and the product gives you a short, ranked list of what to bring — with a plain explanation of what the agent has to guess if you skip each one. Telling us something does not exist is a real answer.

Tasks

A task is one piece of work queued for the project — written by you or split out by the agent from a larger request — and it is the record of what is planned rather than the mechanism that does the work.

Feature briefs

A feature brief is the agent’s written proposal for a change large enough that building it straight away would be a guess: what it understood, what it plans to do, and what it needs you to decide.

Project history

Every project keeps a record of the agent sessions, releases and review runs behind it, so what changed, when, and at what cost is answerable after the fact rather than reconstructed.

Ownership zones explained

Every part of your codebase carries one of three markers — //!AUTO (the agent owns it), //!SAFE (the agent proposes, you approve), //!DEV (humans only) — written in the source itself, so AI permissions are code you can read and change.

Editing generated code

Edit anything you like — your changes are checkpointed as their own version before the next agent turn — but code you intend to keep hand-writing belongs in //!DEV or //!SAFE, or a later turn may rewrite it.

When the agent stops

A stopped turn is usually governance working as designed: the agent needs an answer, needs your approval on a //!SAFE diff, hit a failing check, or reached an action that requires a verified expert's verdict.

Design

Design settles what the app is: screens, flows, data model and architecture, running against mocked data with nothing deployed, so changing your mind is nearly free.

Build

Build replaces mocked data with a real backend, database and integrations on the Build Server, so the app works end-to-end before anyone pays for production infrastructure.

Rehearsal

Rehearsal runs the finished app on the Rehearsal Server, where you drill the data migration, walk the real journeys, and check security and running cost while it is still safe to be wrong.

Live

Live is production: the new system takes over, the old one is retired, and the same governed workflow keeps shipping changes with versioning, monitoring and backups behind them.

Promoting between stages

Promotion runs a readiness gate — tests, health, architecture and stage-specific checks — that can pass, pass with warnings, or block; it is a check on the project, not a deploy button.

The servers

A project runs on up to four servers — Design, Build, Rehearsal and Live — each created when the project reaches the stage it serves. The Design Server is free and serves your mocked preview; the other three are real infrastructure, monitored and backed up, and drawing from the wallet while they exist.

Wallet and top-ups

Sapilon runs on one prepaid wallet that AI points, AWS hosting and expert time all draw from; there is no subscription and no seats, and the balance cannot go negative.

Points

A point is Sapilon's fixed-price unit of AI usage with one published rate, so an interaction costs a predictable whole number instead of a token bill you can only understand afterwards.

What a turn costs

Every agent turn reports what it cost in points and why, so cost is something you read after each turn rather than discover at the end of the month.

Expert work pricing

Expert work is priced by job size — Small (under 30 minutes) and Medium (about an hour) at fixed prices, Large billed hourly against an estimate and a cap you approve before anything starts.