Skip to content
Abada Platform

Quickstart

One command installs a complete local platform: PostgreSQL, the Engine, Studio, the Agent Worker and a development Keycloak, behind a local HTTP edge. You need Docker with Compose v2.

Install

Section titled “Install”
Terminal window
mkdir abada-quickstart && cd abada-quickstart
curl -fsSL https://install.abadaplatform.com/install.sh | bash

The installer does not ask for anything. To bake in a Gemini API key, pass it as ABADA_LLM_GEMINI_API_KEY to bash; the installer then checks that it can call gemini-3.6-flash and writes it only to the local, untracked .env.dev (mode 0600). Otherwise add any provider later in Studio → Settings → AI Providers, where keys are stored encrypted.

When it finishes, the installer sends one anonymous report (version, system, outcome; your country and region come from the request, and your address is not stored). To turn it off, run curl … | ABADA_USAGE_PING=0 bash or set DO_NOT_TRACK=1. See Privacy and the install report.

On Windows (PowerShell):

Terminal window
New-Item -ItemType Directory abada-quickstart
Set-Location abada-quickstart
Invoke-WebRequest https://raw.githubusercontent.com/bashizip/abada-platform/main/release/quickstart.ps1 -OutFile quickstart.ps1
.\quickstart.ps1

Images are published for linux/amd64 and linux/arm64.

Sign in

Section titled “Sign in”

When the installer finishes it prints the URLs and accounts:

What Where
Studio http://studio.localhost (or http://localhost:5605)
Engine information http://api.localhost/api/v1/info
Keycloak (development only) http://keycloak.localhost

Sign in to Studio as alice / alice. A second account, bob / bob, is useful to play the reviewer. These are local development credentials; never reuse them elsewhere.

Studio’s header has four views — Canvas, Task Inbox, Operations and Insight — plus Admin for members of abada-admin.

Run the starter

Section titled “Run the starter”

On first sign-in, Studio creates the Abada Starter project and deploys the AI Lead Triage process. Open Deploy & Start: it offers HIGH, MEDIUM and LOW sample inputs.

  • A HIGH lead creates a review task for bob: sign in as bob, open Task Inbox, claim the task and complete its form.
  • The four LOW samples action feeds the Insight Engine, which then proposes an improvement for a human to review. The model calls run one after another with a quota-safe delay; allow about five minutes.

Check and stop

Section titled “Check and stop”
Terminal window
curl -fsS http://api.localhost/api/actuator/health/readiness
cd abada-platform
./release/abada-platform status dev
./release/abada-platform down dev

Next

Section titled “Next”
  1. Build your own process step by step in the tutorial.
  2. Learn the language in Author processes in APL.
  3. Choose a deployment for a team or a server in Choose a deployment.