Fastex · Desarrollo IT · Soluciones

From an idea to running software.

Describe what you want. The factory plans, builds, tests and runs it — gated on docker compose, shipped to your own Harbor registry and servers, with a human in control at every meaningful step.

The factory, in 3 moves
01
Describe

A business owner explains the need

02
Build & test

Automatic — gated on docker compose

03
Run & deploy

Harbor + your servers, via SSH

Supervised use only — every result must be reviewed.
(01) The challenge

Good ideas wait too long for code

The cost of scarce engineering capacity — felt on every project.

01

Ideas wait in line

Business ideas wait months for scarce engineering capacity.

02

Slow & costly delivery

Delivery is slow and inconsistent across teams and projects.

03

Re-inventing the basics

Every project re-solves the same setup, quality and deployment problems.

04

Talent on plumbing

Senior engineers spend time on plumbing instead of high-value work.

(02) The idea

An AI-powered Software Factory

Describe what you want — the factory builds, tests and runs it.

🤖

AI-powered

A thin layer that drives today's best AI coding agents end to end — no bespoke orchestrator.

📦

Your registry, your servers

Build to a Harbor registry and deploy by SSH with Docker Compose — your own infrastructure, you own the code.

🧭

You stay in control

A human approves every meaningful step — speed without losing the wheel.

(03) End-to-end workflow

The full journey — click any step

A human stays in control at the GATE ✓ and HUMAN checkpoints. The rest runs autonomously.

01 · Describe the need

A non-technical business owner opens Claude Cowork or the GitHub Copilot app and describes the idea in plain language. No specs, no jargon — just the outcome they want.

Human-ledNo code

↑ Steps marked HUMAN or GATE ✓ pause for a person. The rest runs autonomously.

(04) Architecture

One rule: the runtime is the engine

A thin orchestration layer over the best AI coding agents — no bespoke orchestrator code. Hover a layer.

runtime

Runtime

The runtime IS the engine. The factory adds no bespoke orchestrator — it drives Claude Code, opencode or the GitHub Copilot CLI directly. All logic lives in skills, shared verbatim across the three.

Two worlds, one source of truth

Non-technical intake and the technical build meet in the repository.

World 1 · Business (intake)
Business person + AuraOne question/turn · upload designs · no jargon
Aura writes the briefYou don't write a line
The repo · source of truth
.factory/intake/brief · intake.yml · assets
FACTORY.md · roadmapobjectives + the app code
Tracker mirrorGitHub Issues / Azure DevOps
World 2 · Factory (build)
/factory-intakevalidate + human sign-off gate
/factory-init → discoverobjectives + seed the roadmap
/factory-buildphase loop · docker-compose gate
(05) Who does what

Built for both teams

🗣️

You describe the outcome

Explain the need and the outcome you want — no technical knowledge required.

🧩

Aura guides you

A jargon-free, one-question-at-a-time interview writes the brief for you.

You validate & decide

Review the requirements at the sign-off gate, then watch the first working slice.

🎛️

You steer the factory

Review and guide each phase, approve outward actions, own quality and release.

⚙️

Run the command flow

intake → init → discover → sync → build → quality → cicd → deploy → status.

🔍

You validate everything

Read, test and analyse every result before it ships. The factory accelerates you — it doesn't replace your judgement.

(06) Get started

Three steps to your first build

Pick a pilot idea with a clear, measurable outcome. Click each step — or open the docs in the repo.

The technical team creates the GitHub repository, connects it to the factory and checks the prerequisites — git · gh · docker and a runtime (Claude Code / opencode / GitHub Copilot CLI). Then install the factory.

A · New project (greenfield)
# create the project folder and enter it (inside WSL)
mkdir mi-proyecto && cd mi-proyecto && git init
# pull the factory and install it into the repo
git clone https://github.com/Fastex-Soluciones-IT/factoria-software.git ~/factoria-software
~/factoria-software/factory/install.sh .
# gh auth login
gh auth login
Read the QUICKSTART →

The functional team shapes the idea with the guided Aura intake — in Claude Cowork or the GitHub Copilot app — defining the goal, who will use it, and the measurable criteria.

  • Open Claude Cowork or your Copilot Space (repo attached)
  • Answer Aura's questions, one at a time
  • Aura produces the brief → .factory/intake/ → /factory-intake (sign-off gate)
Read the QUICKSTART →

Inside the repo, launch your runtime and let the agent work — it reads the repo, sets itself up and builds, validating every result.

Launch the agent in the repo
cd mi-proyecto
claude   # o:  opencode
# …and inside the agent, type:
guíame
Read the README →
(07) Where it fits at Fastex

Your standards, your registry, your servers

Build

  • Starterkits: FastAPI · OpenSaaS · Frappe
  • SDD engine: gsd · spec-kit · openspec
  • CocoIndex code intelligence (ccc)

Quality

  • GitHub / Azure DevOps as control plane
  • Tests · coverage · minimalism (ponytail)
  • SonarQube

Run & deploy

  • Docker Compose verification gate
  • Harbor registry (your own)
  • Deploy by SSH to your servers

🔐 Secrets in .env.cicd → GitHub Variables/Secrets — never in config.

(08) Responsible use

A power tool — supervised use only

The Software Factory accelerates professional engineers. It does not replace them. Every result must be reviewed, tested and analysed by a qualified developer before it is trusted or shipped.

[✓]Pro-coders supervise. AI-generated code is a draft, not a delivery. A human engineer owns it.
[✓]Validate manually. Read, test and analyse every output — logic, security, performance and data integrity.
[✓]Human-in-the-loop gates. Outward actions (publishing, deploying) always need explicit human approval.
[✓]Secrets stay out. Tokens, SSH keys and the Harbor password live in git-ignored env files; never committed.

Turn an idea into running software

Pick a pilot, run the guided intake, and let the factory build the first working slice — with your team in control.